Branch data Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * heapam_handler.c
4 : : * heap table access method code
5 : : *
6 : : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : : * Portions Copyright (c) 1994, Regents of the University of California
8 : : *
9 : : *
10 : : * IDENTIFICATION
11 : : * src/backend/access/heap/heapam_handler.c
12 : : *
13 : : *
14 : : * NOTES
15 : : * This files wires up the lower level heapam.c et al routines with the
16 : : * tableam abstraction.
17 : : *
18 : : *-------------------------------------------------------------------------
19 : : */
20 : : #include "postgres.h"
21 : :
22 : : #include "access/genam.h"
23 : : #include "access/heapam.h"
24 : : #include "access/heaptoast.h"
25 : : #include "access/multixact.h"
26 : : #include "access/rewriteheap.h"
27 : : #include "access/syncscan.h"
28 : : #include "access/tableam.h"
29 : : #include "access/tsmapi.h"
30 : : #include "access/visibilitymap.h"
31 : : #include "access/xact.h"
32 : : #include "catalog/catalog.h"
33 : : #include "catalog/index.h"
34 : : #include "catalog/storage.h"
35 : : #include "catalog/storage_xlog.h"
36 : : #include "commands/progress.h"
37 : : #include "executor/executor.h"
38 : : #include "miscadmin.h"
39 : : #include "pgstat.h"
40 : : #include "storage/bufmgr.h"
41 : : #include "storage/bufpage.h"
42 : : #include "storage/lmgr.h"
43 : : #include "storage/predicate.h"
44 : : #include "storage/procarray.h"
45 : : #include "storage/smgr.h"
46 : : #include "utils/builtins.h"
47 : : #include "utils/rel.h"
48 : :
49 : : static void reform_and_rewrite_tuple(HeapTuple tuple,
50 : : Relation OldHeap, Relation NewHeap,
51 : : Datum *values, bool *isnull, RewriteState rwstate);
52 : :
53 : : static bool SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
54 : : HeapTuple tuple,
55 : : OffsetNumber tupoffset);
56 : :
57 : : static BlockNumber heapam_scan_get_blocks_done(HeapScanDesc hscan);
58 : :
59 : : static bool BitmapHeapScanNextBlock(TableScanDesc scan,
60 : : bool *recheck,
61 : : uint64 *lossy_pages, uint64 *exact_pages);
62 : :
63 : :
64 : : /* ------------------------------------------------------------------------
65 : : * Slot related callbacks for heap AM
66 : : * ------------------------------------------------------------------------
67 : : */
68 : :
69 : : static const TupleTableSlotOps *
70 : 3620701 : heapam_slot_callbacks(Relation relation)
71 : : {
72 : 3620701 : return &TTSOpsBufferHeapTuple;
73 : : }
74 : :
75 : :
76 : : /* ------------------------------------------------------------------------
77 : : * Index Scan Callbacks for heap AM
78 : : * ------------------------------------------------------------------------
79 : : */
80 : :
81 : : static IndexFetchTableData *
82 : 2840726 : heapam_index_fetch_begin(Relation rel)
83 : : {
84 : 2840726 : IndexFetchHeapData *hscan = palloc0_object(IndexFetchHeapData);
85 : :
86 : 2840726 : hscan->xs_base.rel = rel;
87 : 2840726 : hscan->xs_cbuf = InvalidBuffer;
88 : :
89 : 5681452 : return &hscan->xs_base;
90 : 2840726 : }
91 : :
92 : : static void
93 : 4412379 : heapam_index_fetch_reset(IndexFetchTableData *scan)
94 : : {
95 : 4412379 : IndexFetchHeapData *hscan = (IndexFetchHeapData *) scan;
96 : :
97 [ + + ]: 4412379 : if (BufferIsValid(hscan->xs_cbuf))
98 : : {
99 : 2575273 : ReleaseBuffer(hscan->xs_cbuf);
100 : 2575273 : hscan->xs_cbuf = InvalidBuffer;
101 : 2575273 : }
102 : 4412379 : }
103 : :
104 : : static void
105 : 2840487 : heapam_index_fetch_end(IndexFetchTableData *scan)
106 : : {
107 : 2840487 : IndexFetchHeapData *hscan = (IndexFetchHeapData *) scan;
108 : :
109 : 2840487 : heapam_index_fetch_reset(scan);
110 : :
111 : 2840487 : pfree(hscan);
112 : 2840487 : }
113 : :
114 : : static bool
115 : 3810502 : heapam_index_fetch_tuple(struct IndexFetchTableData *scan,
116 : : ItemPointer tid,
117 : : Snapshot snapshot,
118 : : TupleTableSlot *slot,
119 : : bool *call_again, bool *all_dead)
120 : : {
121 : 3810502 : IndexFetchHeapData *hscan = (IndexFetchHeapData *) scan;
122 : 3810502 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
123 : 3810502 : bool got_heap_tuple;
124 : :
125 [ + - ]: 3810502 : Assert(TTS_IS_BUFFERTUPLE(slot));
126 : :
127 : : /* We can skip the buffer-switching logic if we're in mid-HOT chain. */
128 [ + + ]: 3810502 : if (!*call_again)
129 : : {
130 : : /* Switch to correct buffer if we don't have it already */
131 : 3806665 : Buffer prev_buf = hscan->xs_cbuf;
132 : :
133 : 7613330 : hscan->xs_cbuf = ReleaseAndReadBuffer(hscan->xs_cbuf,
134 : 3806665 : hscan->xs_base.rel,
135 : 3806665 : ItemPointerGetBlockNumber(tid));
136 : :
137 : : /*
138 : : * Prune page, but only if we weren't already on this page
139 : : */
140 [ + + ]: 3806665 : if (prev_buf != hscan->xs_cbuf)
141 : 2927569 : heap_page_prune_opt(hscan->xs_base.rel, hscan->xs_cbuf);
142 : 3806665 : }
143 : :
144 : : /* Obtain share-lock on the buffer so we can examine visibility */
145 : 3810502 : LockBuffer(hscan->xs_cbuf, BUFFER_LOCK_SHARE);
146 : 7621004 : got_heap_tuple = heap_hot_search_buffer(tid,
147 : 3810502 : hscan->xs_base.rel,
148 : 3810502 : hscan->xs_cbuf,
149 : 3810502 : snapshot,
150 : 3810502 : &bslot->base.tupdata,
151 : 3810502 : all_dead,
152 : 3810502 : !*call_again);
153 : 3810502 : bslot->base.tupdata.t_self = *tid;
154 : 3810502 : LockBuffer(hscan->xs_cbuf, BUFFER_LOCK_UNLOCK);
155 : :
156 [ + + ]: 3810502 : if (got_heap_tuple)
157 : : {
158 : : /*
159 : : * Only in a non-MVCC snapshot can more than one member of the HOT
160 : : * chain be visible.
161 : : */
162 [ + + ]: 1823013 : *call_again = !IsMVCCSnapshot(snapshot);
163 : :
164 : 1823013 : slot->tts_tableOid = RelationGetRelid(scan->rel);
165 : 1823013 : ExecStoreBufferHeapTuple(&bslot->base.tupdata, slot, hscan->xs_cbuf);
166 : 1823013 : }
167 : : else
168 : : {
169 : : /* We've reached the end of the HOT chain. */
170 : 1987489 : *call_again = false;
171 : : }
172 : :
173 : 7621004 : return got_heap_tuple;
174 : 3810502 : }
175 : :
176 : :
177 : : /* ------------------------------------------------------------------------
178 : : * Callbacks for non-modifying operations on individual tuples for heap AM
179 : : * ------------------------------------------------------------------------
180 : : */
181 : :
182 : : static bool
183 : 411926 : heapam_fetch_row_version(Relation relation,
184 : : ItemPointer tid,
185 : : Snapshot snapshot,
186 : : TupleTableSlot *slot)
187 : : {
188 : 411926 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
189 : 411926 : Buffer buffer;
190 : :
191 [ + - ]: 411926 : Assert(TTS_IS_BUFFERTUPLE(slot));
192 : :
193 : 411926 : bslot->base.tupdata.t_self = *tid;
194 [ + + ]: 411926 : if (heap_fetch(relation, snapshot, &bslot->base.tupdata, &buffer, false))
195 : : {
196 : : /* store in slot, transferring existing pin */
197 : 411920 : ExecStorePinnedBufferHeapTuple(&bslot->base.tupdata, slot, buffer);
198 : 411920 : slot->tts_tableOid = RelationGetRelid(relation);
199 : :
200 : 411920 : return true;
201 : : }
202 : :
203 : 6 : return false;
204 : 411926 : }
205 : :
206 : : static bool
207 : 76 : heapam_tuple_tid_valid(TableScanDesc scan, ItemPointer tid)
208 : : {
209 : 76 : HeapScanDesc hscan = (HeapScanDesc) scan;
210 : :
211 [ + + ]: 149 : return ItemPointerIsValid(tid) &&
212 : 73 : ItemPointerGetBlockNumber(tid) < hscan->rs_nblocks;
213 : 76 : }
214 : :
215 : : static bool
216 : 427603 : heapam_tuple_satisfies_snapshot(Relation rel, TupleTableSlot *slot,
217 : : Snapshot snapshot)
218 : : {
219 : 427603 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
220 : 427603 : bool res;
221 : :
222 [ + - ]: 427603 : Assert(TTS_IS_BUFFERTUPLE(slot));
223 [ + - ]: 427603 : Assert(BufferIsValid(bslot->buffer));
224 : :
225 : : /*
226 : : * We need buffer pin and lock to call HeapTupleSatisfiesVisibility.
227 : : * Caller should be holding pin, but not lock.
228 : : */
229 : 427603 : LockBuffer(bslot->buffer, BUFFER_LOCK_SHARE);
230 : 855206 : res = HeapTupleSatisfiesVisibility(bslot->base.tuple, snapshot,
231 : 427603 : bslot->buffer);
232 : 427603 : LockBuffer(bslot->buffer, BUFFER_LOCK_UNLOCK);
233 : :
234 : 855206 : return res;
235 : 427603 : }
236 : :
237 : :
238 : : /* ----------------------------------------------------------------------------
239 : : * Functions for manipulations of physical tuples for heap AM.
240 : : * ----------------------------------------------------------------------------
241 : : */
242 : :
243 : : static void
244 : 1621141 : heapam_tuple_insert(Relation relation, TupleTableSlot *slot, CommandId cid,
245 : : int options, BulkInsertState bistate)
246 : : {
247 : 1621141 : bool shouldFree = true;
248 : 1621141 : HeapTuple tuple = ExecFetchSlotHeapTuple(slot, true, &shouldFree);
249 : :
250 : : /* Update the tuple with table oid */
251 : 1621141 : slot->tts_tableOid = RelationGetRelid(relation);
252 : 1621141 : tuple->t_tableOid = slot->tts_tableOid;
253 : :
254 : : /* Perform the insertion, and copy the resulting ItemPointer */
255 : 1621141 : heap_insert(relation, tuple, cid, options, bistate);
256 : 1621141 : ItemPointerCopy(&tuple->t_self, &slot->tts_tid);
257 : :
258 [ + + ]: 1621141 : if (shouldFree)
259 : 376164 : pfree(tuple);
260 : 1621141 : }
261 : :
262 : : static void
263 : 87 : heapam_tuple_insert_speculative(Relation relation, TupleTableSlot *slot,
264 : : CommandId cid, int options,
265 : : BulkInsertState bistate, uint32 specToken)
266 : : {
267 : 87 : bool shouldFree = true;
268 : 87 : HeapTuple tuple = ExecFetchSlotHeapTuple(slot, true, &shouldFree);
269 : :
270 : : /* Update the tuple with table oid */
271 : 87 : slot->tts_tableOid = RelationGetRelid(relation);
272 : 87 : tuple->t_tableOid = slot->tts_tableOid;
273 : :
274 : 87 : HeapTupleHeaderSetSpeculativeToken(tuple->t_data, specToken);
275 : 87 : options |= HEAP_INSERT_SPECULATIVE;
276 : :
277 : : /* Perform the insertion, and copy the resulting ItemPointer */
278 : 87 : heap_insert(relation, tuple, cid, options, bistate);
279 : 87 : ItemPointerCopy(&tuple->t_self, &slot->tts_tid);
280 : :
281 [ + + ]: 87 : if (shouldFree)
282 : 9 : pfree(tuple);
283 : 87 : }
284 : :
285 : : static void
286 : 86 : heapam_tuple_complete_speculative(Relation relation, TupleTableSlot *slot,
287 : : uint32 specToken, bool succeeded)
288 : : {
289 : 86 : bool shouldFree = true;
290 : 86 : HeapTuple tuple = ExecFetchSlotHeapTuple(slot, true, &shouldFree);
291 : :
292 : : /* adjust the tuple's state accordingly */
293 [ + - ]: 86 : if (succeeded)
294 : 86 : heap_finish_speculative(relation, &slot->tts_tid);
295 : : else
296 : 0 : heap_abort_speculative(relation, &slot->tts_tid);
297 : :
298 [ + + ]: 86 : if (shouldFree)
299 : 9 : pfree(tuple);
300 : 86 : }
301 : :
302 : : static TM_Result
303 : 184984 : heapam_tuple_delete(Relation relation, ItemPointer tid, CommandId cid,
304 : : Snapshot snapshot, Snapshot crosscheck, bool wait,
305 : : TM_FailureData *tmfd, bool changingPart)
306 : : {
307 : : /*
308 : : * Currently Deleting of index tuples are handled at vacuum, in case if
309 : : * the storage itself is cleaning the dead tuples by itself, it is the
310 : : * time to call the index tuple deletion also.
311 : : */
312 : 184984 : return heap_delete(relation, tid, cid, crosscheck, wait, tmfd, changingPart);
313 : : }
314 : :
315 : :
316 : : static TM_Result
317 : 6600 : heapam_tuple_update(Relation relation, ItemPointer otid, TupleTableSlot *slot,
318 : : CommandId cid, Snapshot snapshot, Snapshot crosscheck,
319 : : bool wait, TM_FailureData *tmfd,
320 : : LockTupleMode *lockmode, TU_UpdateIndexes *update_indexes)
321 : : {
322 : 6600 : bool shouldFree = true;
323 : 6600 : HeapTuple tuple = ExecFetchSlotHeapTuple(slot, true, &shouldFree);
324 : 6600 : TM_Result result;
325 : :
326 : : /* Update the tuple with table oid */
327 : 6600 : slot->tts_tableOid = RelationGetRelid(relation);
328 : 6600 : tuple->t_tableOid = slot->tts_tableOid;
329 : :
330 : 13200 : result = heap_update(relation, otid, tuple, cid, crosscheck, wait,
331 : 6600 : tmfd, lockmode, update_indexes);
332 : 6600 : ItemPointerCopy(&tuple->t_self, &slot->tts_tid);
333 : :
334 : : /*
335 : : * Decide whether new index entries are needed for the tuple
336 : : *
337 : : * Note: heap_update returns the tid (location) of the new tuple in the
338 : : * t_self field.
339 : : *
340 : : * If the update is not HOT, we must update all indexes. If the update is
341 : : * HOT, it could be that we updated summarized columns, so we either
342 : : * update only summarized indexes, or none at all.
343 : : */
344 [ + + ]: 6600 : if (result != TM_Ok)
345 : : {
346 [ + - ]: 17 : Assert(*update_indexes == TU_None);
347 : 17 : *update_indexes = TU_None;
348 : 17 : }
349 [ + + ]: 6583 : else if (!HeapTupleIsHeapOnly(tuple))
350 [ + - ]: 4366 : Assert(*update_indexes == TU_All);
351 : : else
352 [ + + + - ]: 2217 : Assert((*update_indexes == TU_Summarizing) ||
353 : : (*update_indexes == TU_None));
354 : :
355 [ + + ]: 6600 : if (shouldFree)
356 : 10 : pfree(tuple);
357 : :
358 : 13200 : return result;
359 : 6600 : }
360 : :
361 : : static TM_Result
362 : 400949 : heapam_tuple_lock(Relation relation, ItemPointer tid, Snapshot snapshot,
363 : : TupleTableSlot *slot, CommandId cid, LockTupleMode mode,
364 : : LockWaitPolicy wait_policy, uint8 flags,
365 : : TM_FailureData *tmfd)
366 : : {
367 : 400949 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
368 : 400949 : TM_Result result;
369 : 400949 : Buffer buffer;
370 : 400949 : HeapTuple tuple = &bslot->base.tupdata;
371 : 400949 : bool follow_updates;
372 : :
373 : 400949 : follow_updates = (flags & TUPLE_LOCK_FLAG_LOCK_UPDATE_IN_PROGRESS) != 0;
374 : 400949 : tmfd->traversed = false;
375 : :
376 [ + - ]: 400949 : Assert(TTS_IS_BUFFERTUPLE(slot));
377 : :
378 : : tuple_lock_retry:
379 : 400949 : tuple->t_self = *tid;
380 : 801898 : result = heap_lock_tuple(relation, tuple, cid, mode, wait_policy,
381 : 400949 : follow_updates, &buffer, tmfd);
382 : :
383 [ - + # # ]: 400949 : if (result == TM_Updated &&
384 : 0 : (flags & TUPLE_LOCK_FLAG_FIND_LAST_VERSION))
385 : : {
386 : : /* Should not encounter speculative tuple on recheck */
387 [ # # ]: 0 : Assert(!HeapTupleHeaderIsSpeculative(tuple->t_data));
388 : :
389 : 0 : ReleaseBuffer(buffer);
390 : :
391 [ # # ]: 0 : if (!ItemPointerEquals(&tmfd->ctid, &tuple->t_self))
392 : : {
393 : 0 : SnapshotData SnapshotDirty;
394 : 0 : TransactionId priorXmax;
395 : :
396 : : /* it was updated, so look at the updated version */
397 : 0 : *tid = tmfd->ctid;
398 : : /* updated row should have xmin matching this xmax */
399 : 0 : priorXmax = tmfd->xmax;
400 : :
401 : : /* signal that a tuple later in the chain is getting locked */
402 : 0 : tmfd->traversed = true;
403 : :
404 : : /*
405 : : * fetch target tuple
406 : : *
407 : : * Loop here to deal with updated or busy tuples
408 : : */
409 : 0 : InitDirtySnapshot(SnapshotDirty);
410 : 0 : for (;;)
411 : : {
412 [ # # ]: 0 : if (ItemPointerIndicatesMovedPartitions(tid))
413 [ # # # # ]: 0 : ereport(ERROR,
414 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
415 : : errmsg("tuple to be locked was already moved to another partition due to concurrent update")));
416 : :
417 : 0 : tuple->t_self = *tid;
418 [ # # ]: 0 : if (heap_fetch(relation, &SnapshotDirty, tuple, &buffer, true))
419 : : {
420 : : /*
421 : : * If xmin isn't what we're expecting, the slot must have
422 : : * been recycled and reused for an unrelated tuple. This
423 : : * implies that the latest version of the row was deleted,
424 : : * so we need do nothing. (Should be safe to examine xmin
425 : : * without getting buffer's content lock. We assume
426 : : * reading a TransactionId to be atomic, and Xmin never
427 : : * changes in an existing tuple, except to invalid or
428 : : * frozen, and neither of those can match priorXmax.)
429 : : */
430 [ # # ]: 0 : if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple->t_data),
431 : : priorXmax))
432 : : {
433 : 0 : ReleaseBuffer(buffer);
434 : 0 : return TM_Deleted;
435 : : }
436 : :
437 : : /* otherwise xmin should not be dirty... */
438 [ # # ]: 0 : if (TransactionIdIsValid(SnapshotDirty.xmin))
439 [ # # # # ]: 0 : ereport(ERROR,
440 : : (errcode(ERRCODE_DATA_CORRUPTED),
441 : : errmsg_internal("t_xmin %u is uncommitted in tuple (%u,%u) to be updated in table \"%s\"",
442 : : SnapshotDirty.xmin,
443 : : ItemPointerGetBlockNumber(&tuple->t_self),
444 : : ItemPointerGetOffsetNumber(&tuple->t_self),
445 : : RelationGetRelationName(relation))));
446 : :
447 : : /*
448 : : * If tuple is being updated by other transaction then we
449 : : * have to wait for its commit/abort, or die trying.
450 : : */
451 [ # # ]: 0 : if (TransactionIdIsValid(SnapshotDirty.xmax))
452 : : {
453 : 0 : ReleaseBuffer(buffer);
454 [ # # # # ]: 0 : switch (wait_policy)
455 : : {
456 : : case LockWaitBlock:
457 : 0 : XactLockTableWait(SnapshotDirty.xmax,
458 : 0 : relation, &tuple->t_self,
459 : : XLTW_FetchUpdated);
460 : 0 : break;
461 : : case LockWaitSkip:
462 [ # # ]: 0 : if (!ConditionalXactLockTableWait(SnapshotDirty.xmax, false))
463 : : /* skip instead of waiting */
464 : 0 : return TM_WouldBlock;
465 : 0 : break;
466 : : case LockWaitError:
467 [ # # ]: 0 : if (!ConditionalXactLockTableWait(SnapshotDirty.xmax, log_lock_failures))
468 [ # # # # ]: 0 : ereport(ERROR,
469 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
470 : : errmsg("could not obtain lock on row in relation \"%s\"",
471 : : RelationGetRelationName(relation))));
472 : 0 : break;
473 : : }
474 : 0 : continue; /* loop back to repeat heap_fetch */
475 : : }
476 : :
477 : : /*
478 : : * If tuple was inserted by our own transaction, we have
479 : : * to check cmin against cid: cmin >= current CID means
480 : : * our command cannot see the tuple, so we should ignore
481 : : * it. Otherwise heap_lock_tuple() will throw an error,
482 : : * and so would any later attempt to update or delete the
483 : : * tuple. (We need not check cmax because
484 : : * HeapTupleSatisfiesDirty will consider a tuple deleted
485 : : * by our transaction dead, regardless of cmax.) We just
486 : : * checked that priorXmax == xmin, so we can test that
487 : : * variable instead of doing HeapTupleHeaderGetXmin again.
488 : : */
489 [ # # # # ]: 0 : if (TransactionIdIsCurrentTransactionId(priorXmax) &&
490 : 0 : HeapTupleHeaderGetCmin(tuple->t_data) >= cid)
491 : : {
492 : 0 : tmfd->xmax = priorXmax;
493 : :
494 : : /*
495 : : * Cmin is the problematic value, so store that. See
496 : : * above.
497 : : */
498 : 0 : tmfd->cmax = HeapTupleHeaderGetCmin(tuple->t_data);
499 : 0 : ReleaseBuffer(buffer);
500 : 0 : return TM_SelfModified;
501 : : }
502 : :
503 : : /*
504 : : * This is a live tuple, so try to lock it again.
505 : : */
506 : 0 : ReleaseBuffer(buffer);
507 : 0 : goto tuple_lock_retry;
508 : : }
509 : :
510 : : /*
511 : : * If the referenced slot was actually empty, the latest
512 : : * version of the row must have been deleted, so we need do
513 : : * nothing.
514 : : */
515 [ # # ]: 0 : if (tuple->t_data == NULL)
516 : : {
517 [ # # ]: 0 : Assert(!BufferIsValid(buffer));
518 : 0 : return TM_Deleted;
519 : : }
520 : :
521 : : /*
522 : : * As above, if xmin isn't what we're expecting, do nothing.
523 : : */
524 [ # # ]: 0 : if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple->t_data),
525 : : priorXmax))
526 : : {
527 : 0 : ReleaseBuffer(buffer);
528 : 0 : return TM_Deleted;
529 : : }
530 : :
531 : : /*
532 : : * If we get here, the tuple was found but failed
533 : : * SnapshotDirty. Assuming the xmin is either a committed xact
534 : : * or our own xact (as it certainly should be if we're trying
535 : : * to modify the tuple), this must mean that the row was
536 : : * updated or deleted by either a committed xact or our own
537 : : * xact. If it was deleted, we can ignore it; if it was
538 : : * updated then chain up to the next version and repeat the
539 : : * whole process.
540 : : *
541 : : * As above, it should be safe to examine xmax and t_ctid
542 : : * without the buffer content lock, because they can't be
543 : : * changing. We'd better hold a buffer pin though.
544 : : */
545 [ # # ]: 0 : if (ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
546 : : {
547 : : /* deleted, so forget about it */
548 : 0 : ReleaseBuffer(buffer);
549 : 0 : return TM_Deleted;
550 : : }
551 : :
552 : : /* updated, so look at the updated row */
553 : 0 : *tid = tuple->t_data->t_ctid;
554 : : /* updated row should have xmin matching this xmax */
555 : 0 : priorXmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
556 : 0 : ReleaseBuffer(buffer);
557 : : /* loop back to fetch next in chain */
558 : : }
559 [ # # ]: 0 : }
560 : : else
561 : : {
562 : : /* tuple was deleted, so give up */
563 : 0 : return TM_Deleted;
564 : : }
565 : : }
566 : :
567 : 400949 : slot->tts_tableOid = RelationGetRelid(relation);
568 : 400949 : tuple->t_tableOid = slot->tts_tableOid;
569 : :
570 : : /* store in slot, transferring existing pin */
571 : 400949 : ExecStorePinnedBufferHeapTuple(tuple, slot, buffer);
572 : :
573 : 400949 : return result;
574 : 400949 : }
575 : :
576 : :
577 : : /* ------------------------------------------------------------------------
578 : : * DDL related callbacks for heap AM.
579 : : * ------------------------------------------------------------------------
580 : : */
581 : :
582 : : static void
583 : 6439 : heapam_relation_set_new_filelocator(Relation rel,
584 : : const RelFileLocator *newrlocator,
585 : : char persistence,
586 : : TransactionId *freezeXid,
587 : : MultiXactId *minmulti)
588 : : {
589 : 6439 : SMgrRelation srel;
590 : :
591 : : /*
592 : : * Initialize to the minimum XID that could put tuples in the table. We
593 : : * know that no xacts older than RecentXmin are still running, so that
594 : : * will do.
595 : : */
596 : 6439 : *freezeXid = RecentXmin;
597 : :
598 : : /*
599 : : * Similarly, initialize the minimum Multixact to the first value that
600 : : * could possibly be stored in tuples in the table. Running transactions
601 : : * could reuse values from their local cache, so we are careful to
602 : : * consider all currently running multis.
603 : : *
604 : : * XXX this could be refined further, but is it worth the hassle?
605 : : */
606 : 6439 : *minmulti = GetOldestMultiXactId();
607 : :
608 : 6439 : srel = RelationCreateStorage(*newrlocator, persistence, true);
609 : :
610 : : /*
611 : : * If required, set up an init fork for an unlogged table so that it can
612 : : * be correctly reinitialized on restart.
613 : : */
614 [ + + ]: 6439 : if (persistence == RELPERSISTENCE_UNLOGGED)
615 : : {
616 [ + + + - ]: 27 : Assert(rel->rd_rel->relkind == RELKIND_RELATION ||
617 : : rel->rd_rel->relkind == RELKIND_TOASTVALUE);
618 : 27 : smgrcreate(srel, INIT_FORKNUM, false);
619 : 27 : log_smgrcreate(newrlocator, INIT_FORKNUM);
620 : 27 : }
621 : :
622 : 6439 : smgrclose(srel);
623 : 6439 : }
624 : :
625 : : static void
626 : 86 : heapam_relation_nontransactional_truncate(Relation rel)
627 : : {
628 : 86 : RelationTruncate(rel, 0);
629 : 86 : }
630 : :
631 : : static void
632 : 14 : heapam_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
633 : : {
634 : 14 : SMgrRelation dstrel;
635 : :
636 : : /*
637 : : * Since we copy the file directly without looking at the shared buffers,
638 : : * we'd better first flush out any pages of the source relation that are
639 : : * in shared buffers. We assume no new changes will be made while we are
640 : : * holding exclusive lock on the rel.
641 : : */
642 : 14 : FlushRelationBuffers(rel);
643 : :
644 : : /*
645 : : * Create and copy all forks of the relation, and schedule unlinking of
646 : : * old physical files.
647 : : *
648 : : * NOTE: any conflict in relfilenumber value will be caught in
649 : : * RelationCreateStorage().
650 : : */
651 : 14 : dstrel = RelationCreateStorage(*newrlocator, rel->rd_rel->relpersistence, true);
652 : :
653 : : /* copy main fork */
654 : 28 : RelationCopyStorage(RelationGetSmgr(rel), dstrel, MAIN_FORKNUM,
655 : 14 : rel->rd_rel->relpersistence);
656 : :
657 : : /* copy those extra forks that exist */
658 [ + + ]: 56 : for (ForkNumber forkNum = MAIN_FORKNUM + 1;
659 : 56 : forkNum <= MAX_FORKNUM; forkNum++)
660 : : {
661 [ + + ]: 42 : if (smgrexists(RelationGetSmgr(rel), forkNum))
662 : : {
663 : 3 : smgrcreate(dstrel, forkNum, false);
664 : :
665 : : /*
666 : : * WAL log creation if the relation is persistent, or this is the
667 : : * init fork of an unlogged relation.
668 : : */
669 [ + + # # ]: 3 : if (RelationIsPermanent(rel) ||
670 [ - + ]: 1 : (rel->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED &&
671 : 0 : forkNum == INIT_FORKNUM))
672 : 2 : log_smgrcreate(newrlocator, forkNum);
673 : 6 : RelationCopyStorage(RelationGetSmgr(rel), dstrel, forkNum,
674 : 3 : rel->rd_rel->relpersistence);
675 : 3 : }
676 : 42 : }
677 : :
678 : :
679 : : /* drop old relation, and close new one */
680 : 14 : RelationDropStorage(rel);
681 : 14 : smgrclose(dstrel);
682 : 14 : }
683 : :
684 : : static void
685 : 54 : heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
686 : : Relation OldIndex, bool use_sort,
687 : : TransactionId OldestXmin,
688 : : TransactionId *xid_cutoff,
689 : : MultiXactId *multi_cutoff,
690 : : double *num_tuples,
691 : : double *tups_vacuumed,
692 : : double *tups_recently_dead)
693 : : {
694 : 54 : RewriteState rwstate;
695 : 54 : IndexScanDesc indexScan;
696 : 54 : TableScanDesc tableScan;
697 : 54 : HeapScanDesc heapScan;
698 : 54 : bool is_system_catalog;
699 : 54 : Tuplesortstate *tuplesort;
700 : 54 : TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
701 : 54 : TupleDesc newTupDesc = RelationGetDescr(NewHeap);
702 : 54 : TupleTableSlot *slot;
703 : 54 : int natts;
704 : 54 : Datum *values;
705 : 54 : bool *isnull;
706 : 54 : BufferHeapTupleTableSlot *hslot;
707 : 54 : BlockNumber prev_cblock = InvalidBlockNumber;
708 : :
709 : : /* Remember if it's a system catalog */
710 : 54 : is_system_catalog = IsSystemRelation(OldHeap);
711 : :
712 : : /*
713 : : * Valid smgr_targblock implies something already wrote to the relation.
714 : : * This may be harmless, but this function hasn't planned for it.
715 : : */
716 [ - + + - ]: 54 : Assert(RelationGetTargetBlock(NewHeap) == InvalidBlockNumber);
717 : :
718 : : /* Preallocate values/isnull arrays */
719 : 54 : natts = newTupDesc->natts;
720 : 54 : values = palloc_array(Datum, natts);
721 : 54 : isnull = palloc_array(bool, natts);
722 : :
723 : : /* Initialize the rewrite operation */
724 : 108 : rwstate = begin_heap_rewrite(OldHeap, NewHeap, OldestXmin, *xid_cutoff,
725 : 54 : *multi_cutoff);
726 : :
727 : :
728 : : /* Set up sorting if wanted */
729 [ + + ]: 54 : if (use_sort)
730 : 32 : tuplesort = tuplesort_begin_cluster(oldTupDesc, OldIndex,
731 : 16 : maintenance_work_mem,
732 : : NULL, TUPLESORT_NONE);
733 : : else
734 : 38 : tuplesort = NULL;
735 : :
736 : : /*
737 : : * Prepare to scan the OldHeap. To ensure we see recently-dead tuples
738 : : * that still need to be copied, we scan with SnapshotAny and use
739 : : * HeapTupleSatisfiesVacuum for the visibility test.
740 : : */
741 [ + + + + ]: 54 : if (OldIndex != NULL && !use_sort)
742 : : {
743 : 11 : const int ci_index[] = {
744 : : PROGRESS_CLUSTER_PHASE,
745 : : PROGRESS_CLUSTER_INDEX_RELID
746 : : };
747 : 11 : int64 ci_val[2];
748 : :
749 : : /* Set phase and OIDOldIndex to columns */
750 : 11 : ci_val[0] = PROGRESS_CLUSTER_PHASE_INDEX_SCAN_HEAP;
751 : 11 : ci_val[1] = RelationGetRelid(OldIndex);
752 : 11 : pgstat_progress_update_multi_param(2, ci_index, ci_val);
753 : :
754 : 11 : tableScan = NULL;
755 : 11 : heapScan = NULL;
756 : 11 : indexScan = index_beginscan(OldHeap, OldIndex, SnapshotAny, NULL, 0, 0);
757 : 11 : index_rescan(indexScan, NULL, 0, NULL, 0);
758 : 11 : }
759 : : else
760 : : {
761 : : /* In scan-and-sort mode and also VACUUM FULL, set phase */
762 : 43 : pgstat_progress_update_param(PROGRESS_CLUSTER_PHASE,
763 : : PROGRESS_CLUSTER_PHASE_SEQ_SCAN_HEAP);
764 : :
765 : 43 : tableScan = table_beginscan(OldHeap, SnapshotAny, 0, (ScanKey) NULL);
766 : 43 : heapScan = (HeapScanDesc) tableScan;
767 : 43 : indexScan = NULL;
768 : :
769 : : /* Set total heap blocks */
770 : 43 : pgstat_progress_update_param(PROGRESS_CLUSTER_TOTAL_HEAP_BLKS,
771 : 43 : heapScan->rs_nblocks);
772 : : }
773 : :
774 : 54 : slot = table_slot_create(OldHeap, NULL);
775 : 54 : hslot = (BufferHeapTupleTableSlot *) slot;
776 : :
777 : : /*
778 : : * Scan through the OldHeap, either in OldIndex order or sequentially;
779 : : * copy each tuple into the NewHeap, or transiently to the tuplesort
780 : : * module. Note that we don't bother sorting dead tuples (they won't get
781 : : * to the new table anyway).
782 : : */
783 : 114953 : for (;;)
784 : : {
785 : 115385 : HeapTuple tuple;
786 : 115385 : Buffer buf;
787 : 115385 : bool isdead;
788 : :
789 [ + - ]: 115385 : CHECK_FOR_INTERRUPTS();
790 : :
791 [ + + ]: 115385 : if (indexScan != NULL)
792 : : {
793 [ + + ]: 27 : if (!index_getnext_slot(indexScan, ForwardScanDirection, slot))
794 : 11 : break;
795 : :
796 : : /* Since we used no scan keys, should never need to recheck */
797 [ + - ]: 16 : if (indexScan->xs_recheck)
798 [ # # # # ]: 0 : elog(ERROR, "CLUSTER does not support lossy index conditions");
799 : 16 : }
800 : : else
801 : : {
802 [ + + ]: 115358 : if (!table_scan_getnextslot(tableScan, ForwardScanDirection, slot))
803 : : {
804 : : /*
805 : : * If the last pages of the scan were empty, we would go to
806 : : * the next phase while heap_blks_scanned != heap_blks_total.
807 : : * Instead, to ensure that heap_blks_scanned is equivalent to
808 : : * heap_blks_total after the table scan phase, this parameter
809 : : * is manually updated to the correct value when the table
810 : : * scan finishes.
811 : : */
812 : 43 : pgstat_progress_update_param(PROGRESS_CLUSTER_HEAP_BLKS_SCANNED,
813 : 43 : heapScan->rs_nblocks);
814 : 43 : break;
815 : : }
816 : :
817 : : /*
818 : : * In scan-and-sort mode and also VACUUM FULL, set heap blocks
819 : : * scanned
820 : : *
821 : : * Note that heapScan may start at an offset and wrap around, i.e.
822 : : * rs_startblock may be >0, and rs_cblock may end with a number
823 : : * below rs_startblock. To prevent showing this wraparound to the
824 : : * user, we offset rs_cblock by rs_startblock (modulo rs_nblocks).
825 : : */
826 [ + + ]: 115315 : if (prev_cblock != heapScan->rs_cblock)
827 : : {
828 : 1526 : pgstat_progress_update_param(PROGRESS_CLUSTER_HEAP_BLKS_SCANNED,
829 : 4578 : (heapScan->rs_cblock +
830 : 3052 : heapScan->rs_nblocks -
831 : 1526 : heapScan->rs_startblock
832 : 1526 : ) % heapScan->rs_nblocks + 1);
833 : 1526 : prev_cblock = heapScan->rs_cblock;
834 : 1526 : }
835 : : }
836 : :
837 : 115331 : tuple = ExecFetchSlotHeapTuple(slot, false, NULL);
838 : 115331 : buf = hslot->buffer;
839 : :
840 : : /*
841 : : * To be able to guarantee that we can set the hint bit, acquire an
842 : : * exclusive lock on the old buffer. We need the hint bits, set in
843 : : * heapam_relation_copy_for_cluster() -> HeapTupleSatisfiesVacuum(),
844 : : * to be set, as otherwise reform_and_rewrite_tuple() ->
845 : : * rewrite_heap_tuple() will get confused. Specifically,
846 : : * rewrite_heap_tuple() checks for HEAP_XMAX_INVALID in the old tuple
847 : : * to determine whether to check the old-to-new mapping hash table.
848 : : *
849 : : * It'd be better if we somehow could avoid setting hint bits on the
850 : : * old page. One reason to use VACUUM FULL are very bloated tables -
851 : : * rewriting most of the old table during VACUUM FULL doesn't exactly
852 : : * help...
853 : : */
854 : 115331 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
855 : :
856 [ + + + + : 115331 : switch (HeapTupleSatisfiesVacuum(tuple, OldestXmin, buf))
+ - ]
857 : : {
858 : : case HEAPTUPLE_DEAD:
859 : : /* Definitely dead */
860 : 432 : isdead = true;
861 : 432 : break;
862 : : case HEAPTUPLE_RECENTLY_DEAD:
863 : 18961 : *tups_recently_dead += 1;
864 : : /* fall through */
865 : : case HEAPTUPLE_LIVE:
866 : : /* Live or recently dead, must copy it */
867 : 114890 : isdead = false;
868 : 114890 : break;
869 : : case HEAPTUPLE_INSERT_IN_PROGRESS:
870 : :
871 : : /*
872 : : * Since we hold exclusive lock on the relation, normally the
873 : : * only way to see this is if it was inserted earlier in our
874 : : * own transaction. However, it can happen in system
875 : : * catalogs, since we tend to release write lock before commit
876 : : * there. Give a warning if neither case applies; but in any
877 : : * case we had better copy it.
878 : : */
879 [ + + + - ]: 4 : if (!is_system_catalog &&
880 : 3 : !TransactionIdIsCurrentTransactionId(HeapTupleHeaderGetXmin(tuple->t_data)))
881 [ # # # # ]: 0 : elog(WARNING, "concurrent insert in progress within table \"%s\"",
882 : : RelationGetRelationName(OldHeap));
883 : : /* treat as live */
884 : 4 : isdead = false;
885 : 4 : break;
886 : : case HEAPTUPLE_DELETE_IN_PROGRESS:
887 : :
888 : : /*
889 : : * Similar situation to INSERT_IN_PROGRESS case.
890 : : */
891 [ + - + - ]: 5 : if (!is_system_catalog &&
892 : 5 : !TransactionIdIsCurrentTransactionId(HeapTupleHeaderGetUpdateXid(tuple->t_data)))
893 [ # # # # ]: 0 : elog(WARNING, "concurrent delete in progress within table \"%s\"",
894 : : RelationGetRelationName(OldHeap));
895 : : /* treat as recently dead */
896 : 5 : *tups_recently_dead += 1;
897 : 5 : isdead = false;
898 : 5 : break;
899 : : default:
900 [ # # # # ]: 0 : elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result");
901 : 0 : isdead = false; /* keep compiler quiet */
902 : 0 : break;
903 : : }
904 : :
905 : 115331 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
906 : :
907 [ + + ]: 115331 : if (isdead)
908 : : {
909 : 432 : *tups_vacuumed += 1;
910 : : /* heap rewrite module still needs to see it... */
911 [ + - ]: 432 : if (rewrite_heap_dead_tuple(rwstate, tuple))
912 : : {
913 : : /* A previous recently-dead tuple is now known dead */
914 : 0 : *tups_vacuumed += 1;
915 : 0 : *tups_recently_dead -= 1;
916 : 0 : }
917 : 432 : continue;
918 : : }
919 : :
920 : 114899 : *num_tuples += 1;
921 [ + + ]: 114899 : if (tuplesort != NULL)
922 : : {
923 : 90357 : tuplesort_putheaptuple(tuplesort, tuple);
924 : :
925 : : /*
926 : : * In scan-and-sort mode, report increase in number of tuples
927 : : * scanned
928 : : */
929 : 90357 : pgstat_progress_update_param(PROGRESS_CLUSTER_HEAP_TUPLES_SCANNED,
930 : 90357 : *num_tuples);
931 : 90357 : }
932 : : else
933 : : {
934 : 24542 : const int ct_index[] = {
935 : : PROGRESS_CLUSTER_HEAP_TUPLES_SCANNED,
936 : : PROGRESS_CLUSTER_HEAP_TUPLES_WRITTEN
937 : : };
938 : 24542 : int64 ct_val[2];
939 : :
940 : 49084 : reform_and_rewrite_tuple(tuple, OldHeap, NewHeap,
941 : 24542 : values, isnull, rwstate);
942 : :
943 : : /*
944 : : * In indexscan mode and also VACUUM FULL, report increase in
945 : : * number of tuples scanned and written
946 : : */
947 : 24542 : ct_val[0] = *num_tuples;
948 : 24542 : ct_val[1] = *num_tuples;
949 : 24542 : pgstat_progress_update_multi_param(2, ct_index, ct_val);
950 : 24542 : }
951 [ + + + ]: 115385 : }
952 : :
953 [ + + ]: 54 : if (indexScan != NULL)
954 : 11 : index_endscan(indexScan);
955 [ + + ]: 54 : if (tableScan != NULL)
956 : 43 : table_endscan(tableScan);
957 [ + - ]: 54 : if (slot)
958 : 54 : ExecDropSingleTupleTableSlot(slot);
959 : :
960 : : /*
961 : : * In scan-and-sort mode, complete the sort, then read out all live tuples
962 : : * from the tuplestore and write them to the new relation.
963 : : */
964 [ + + ]: 54 : if (tuplesort != NULL)
965 : : {
966 : 16 : double n_tuples = 0;
967 : :
968 : : /* Report that we are now sorting tuples */
969 : 16 : pgstat_progress_update_param(PROGRESS_CLUSTER_PHASE,
970 : : PROGRESS_CLUSTER_PHASE_SORT_TUPLES);
971 : :
972 : 16 : tuplesort_performsort(tuplesort);
973 : :
974 : : /* Report that we are now writing new heap */
975 : 16 : pgstat_progress_update_param(PROGRESS_CLUSTER_PHASE,
976 : : PROGRESS_CLUSTER_PHASE_WRITE_NEW_HEAP);
977 : :
978 : 90373 : for (;;)
979 : : {
980 : 90373 : HeapTuple tuple;
981 : :
982 [ - + ]: 90373 : CHECK_FOR_INTERRUPTS();
983 : :
984 : 90373 : tuple = tuplesort_getheaptuple(tuplesort, true);
985 [ + + ]: 90373 : if (tuple == NULL)
986 : 16 : break;
987 : :
988 : 90357 : n_tuples += 1;
989 : 180714 : reform_and_rewrite_tuple(tuple,
990 : 90357 : OldHeap, NewHeap,
991 : 90357 : values, isnull,
992 : 90357 : rwstate);
993 : : /* Report n_tuples */
994 : 90357 : pgstat_progress_update_param(PROGRESS_CLUSTER_HEAP_TUPLES_WRITTEN,
995 : 90357 : n_tuples);
996 [ + + ]: 90373 : }
997 : :
998 : 16 : tuplesort_end(tuplesort);
999 : 16 : }
1000 : :
1001 : : /* Write out any remaining tuples, and fsync if needed */
1002 : 54 : end_heap_rewrite(rwstate);
1003 : :
1004 : : /* Clean up */
1005 : 54 : pfree(values);
1006 : 54 : pfree(isnull);
1007 : 54 : }
1008 : :
1009 : : /*
1010 : : * Prepare to analyze the next block in the read stream. Returns false if
1011 : : * the stream is exhausted and true otherwise. The scan must have been started
1012 : : * with SO_TYPE_ANALYZE option.
1013 : : *
1014 : : * This routine holds a buffer pin and lock on the heap page. They are held
1015 : : * until heapam_scan_analyze_next_tuple() returns false. That is until all the
1016 : : * items of the heap page are analyzed.
1017 : : */
1018 : : static bool
1019 : 11697 : heapam_scan_analyze_next_block(TableScanDesc scan, ReadStream *stream)
1020 : : {
1021 : 11697 : HeapScanDesc hscan = (HeapScanDesc) scan;
1022 : :
1023 : : /*
1024 : : * We must maintain a pin on the target page's buffer to ensure that
1025 : : * concurrent activity - e.g. HOT pruning - doesn't delete tuples out from
1026 : : * under us. It comes from the stream already pinned. We also choose to
1027 : : * hold sharelock on the buffer throughout --- we could release and
1028 : : * re-acquire sharelock for each tuple, but since we aren't doing much
1029 : : * work per tuple, the extra lock traffic is probably better avoided.
1030 : : */
1031 : 11697 : hscan->rs_cbuf = read_stream_next_buffer(stream, NULL);
1032 [ + + ]: 11697 : if (!BufferIsValid(hscan->rs_cbuf))
1033 : 1000 : return false;
1034 : :
1035 : 10697 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
1036 : :
1037 : 10697 : hscan->rs_cblock = BufferGetBlockNumber(hscan->rs_cbuf);
1038 : 10697 : hscan->rs_cindex = FirstOffsetNumber;
1039 : 10697 : return true;
1040 : 11697 : }
1041 : :
1042 : : static bool
1043 : 1374945 : heapam_scan_analyze_next_tuple(TableScanDesc scan, TransactionId OldestXmin,
1044 : : double *liverows, double *deadrows,
1045 : : TupleTableSlot *slot)
1046 : : {
1047 : 1374945 : HeapScanDesc hscan = (HeapScanDesc) scan;
1048 : 1374945 : Page targpage;
1049 : 1374945 : OffsetNumber maxoffset;
1050 : 1374945 : BufferHeapTupleTableSlot *hslot;
1051 : :
1052 [ + - ]: 1374945 : Assert(TTS_IS_BUFFERTUPLE(slot));
1053 : :
1054 : 1374945 : hslot = (BufferHeapTupleTableSlot *) slot;
1055 : 1374945 : targpage = BufferGetPage(hscan->rs_cbuf);
1056 : 1374945 : maxoffset = PageGetMaxOffsetNumber(targpage);
1057 : :
1058 : : /* Inner loop over all tuples on the selected page */
1059 [ + + ]: 1413409 : for (; hscan->rs_cindex <= maxoffset; hscan->rs_cindex++)
1060 : : {
1061 : 1402712 : ItemId itemid;
1062 : 1402712 : HeapTuple targtuple = &hslot->base.tupdata;
1063 : 1402712 : bool sample_it = false;
1064 : :
1065 : 1402712 : itemid = PageGetItemId(targpage, hscan->rs_cindex);
1066 : :
1067 : : /*
1068 : : * We ignore unused and redirect line pointers. DEAD line pointers
1069 : : * should be counted as dead, because we need vacuum to run to get rid
1070 : : * of them. Note that this rule agrees with the way that
1071 : : * heap_page_prune_and_freeze() counts things.
1072 : : */
1073 [ + + ]: 1402712 : if (!ItemIdIsNormal(itemid))
1074 : : {
1075 [ + + ]: 4749 : if (ItemIdIsDead(itemid))
1076 : 4117 : *deadrows += 1;
1077 : 4749 : continue;
1078 : : }
1079 : :
1080 : 1397963 : ItemPointerSet(&targtuple->t_self, hscan->rs_cblock, hscan->rs_cindex);
1081 : :
1082 : 1397963 : targtuple->t_tableOid = RelationGetRelid(scan->rs_rd);
1083 : 1397963 : targtuple->t_data = (HeapTupleHeader) PageGetItem(targpage, itemid);
1084 : 1397963 : targtuple->t_len = ItemIdGetLength(itemid);
1085 : :
1086 [ - + + + : 2795926 : switch (HeapTupleSatisfiesVacuum(targtuple, OldestXmin,
+ - + + +
+ ]
1087 : 1397963 : hscan->rs_cbuf))
1088 : : {
1089 : : case HEAPTUPLE_LIVE:
1090 : 1311211 : sample_it = true;
1091 : 1311211 : *liverows += 1;
1092 : 1311211 : break;
1093 : :
1094 : : case HEAPTUPLE_DEAD:
1095 : : case HEAPTUPLE_RECENTLY_DEAD:
1096 : : /* Count dead and recently-dead rows */
1097 : 33429 : *deadrows += 1;
1098 : 33429 : break;
1099 : :
1100 : : case HEAPTUPLE_INSERT_IN_PROGRESS:
1101 : :
1102 : : /*
1103 : : * Insert-in-progress rows are not counted. We assume that
1104 : : * when the inserting transaction commits or aborts, it will
1105 : : * send a stats message to increment the proper count. This
1106 : : * works right only if that transaction ends after we finish
1107 : : * analyzing the table; if things happen in the other order,
1108 : : * its stats update will be overwritten by ours. However, the
1109 : : * error will be large only if the other transaction runs long
1110 : : * enough to insert many tuples, so assuming it will finish
1111 : : * after us is the safer option.
1112 : : *
1113 : : * A special case is that the inserting transaction might be
1114 : : * our own. In this case we should count and sample the row,
1115 : : * to accommodate users who load a table and analyze it in one
1116 : : * transaction. (pgstat_report_analyze has to adjust the
1117 : : * numbers we report to the cumulative stats system to make
1118 : : * this come out right.)
1119 : : */
1120 [ - + ]: 53037 : if (TransactionIdIsCurrentTransactionId(HeapTupleHeaderGetXmin(targtuple->t_data)))
1121 : : {
1122 : 53037 : sample_it = true;
1123 : 53037 : *liverows += 1;
1124 : 53037 : }
1125 : 53037 : break;
1126 : :
1127 : : case HEAPTUPLE_DELETE_IN_PROGRESS:
1128 : :
1129 : : /*
1130 : : * We count and sample delete-in-progress rows the same as
1131 : : * live ones, so that the stats counters come out right if the
1132 : : * deleting transaction commits after us, per the same
1133 : : * reasoning given above.
1134 : : *
1135 : : * If the delete was done by our own transaction, however, we
1136 : : * must count the row as dead to make pgstat_report_analyze's
1137 : : * stats adjustments come out right. (Note: this works out
1138 : : * properly when the row was both inserted and deleted in our
1139 : : * xact.)
1140 : : *
1141 : : * The net effect of these choices is that we act as though an
1142 : : * IN_PROGRESS transaction hasn't happened yet, except if it
1143 : : * is our own transaction, which we assume has happened.
1144 : : *
1145 : : * This approach ensures that we behave sanely if we see both
1146 : : * the pre-image and post-image rows for a row being updated
1147 : : * by a concurrent transaction: we will sample the pre-image
1148 : : * but not the post-image. We also get sane results if the
1149 : : * concurrent transaction never commits.
1150 : : */
1151 [ + - ]: 286 : if (TransactionIdIsCurrentTransactionId(HeapTupleHeaderGetUpdateXid(targtuple->t_data)))
1152 : 286 : *deadrows += 1;
1153 : : else
1154 : : {
1155 : 0 : sample_it = true;
1156 : 0 : *liverows += 1;
1157 : : }
1158 : 286 : break;
1159 : :
1160 : : default:
1161 [ # # # # ]: 0 : elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result");
1162 : 0 : break;
1163 : : }
1164 : :
1165 [ + + ]: 1397963 : if (sample_it)
1166 : : {
1167 : 1364248 : ExecStoreBufferHeapTuple(targtuple, slot, hscan->rs_cbuf);
1168 : 1364248 : hscan->rs_cindex++;
1169 : :
1170 : : /* note that we leave the buffer locked here! */
1171 : 1364248 : return true;
1172 : : }
1173 [ + + + ]: 1402712 : }
1174 : :
1175 : : /* Now release the lock and pin on the page */
1176 : 10697 : UnlockReleaseBuffer(hscan->rs_cbuf);
1177 : 10697 : hscan->rs_cbuf = InvalidBuffer;
1178 : :
1179 : : /* also prevent old slot contents from having pin on page */
1180 : 10697 : ExecClearTuple(slot);
1181 : :
1182 : 10697 : return false;
1183 : 1374945 : }
1184 : :
1185 : : static double
1186 : 4057 : heapam_index_build_range_scan(Relation heapRelation,
1187 : : Relation indexRelation,
1188 : : IndexInfo *indexInfo,
1189 : : bool allow_sync,
1190 : : bool anyvisible,
1191 : : bool progress,
1192 : : BlockNumber start_blockno,
1193 : : BlockNumber numblocks,
1194 : : IndexBuildCallback callback,
1195 : : void *callback_state,
1196 : : TableScanDesc scan)
1197 : : {
1198 : 4057 : HeapScanDesc hscan;
1199 : 4057 : bool is_system_catalog;
1200 : 4057 : bool checking_uniqueness;
1201 : 4057 : HeapTuple heapTuple;
1202 : 4057 : Datum values[INDEX_MAX_KEYS];
1203 : 4057 : bool isnull[INDEX_MAX_KEYS];
1204 : 4057 : double reltuples;
1205 : 4057 : ExprState *predicate;
1206 : 4057 : TupleTableSlot *slot;
1207 : 4057 : EState *estate;
1208 : 4057 : ExprContext *econtext;
1209 : 4057 : Snapshot snapshot;
1210 : 4057 : bool need_unregister_snapshot = false;
1211 : 4057 : TransactionId OldestXmin;
1212 : 4057 : BlockNumber previous_blkno = InvalidBlockNumber;
1213 : 4057 : BlockNumber root_blkno = InvalidBlockNumber;
1214 : 4057 : OffsetNumber root_offsets[MaxHeapTuplesPerPage];
1215 : :
1216 : : /*
1217 : : * sanity checks
1218 : : */
1219 [ + - ]: 4057 : Assert(OidIsValid(indexRelation->rd_rel->relam));
1220 : :
1221 : : /* Remember if it's a system catalog */
1222 : 4057 : is_system_catalog = IsSystemRelation(heapRelation);
1223 : :
1224 : : /* See whether we're verifying uniqueness/exclusion properties */
1225 [ + + ]: 4057 : checking_uniqueness = (indexInfo->ii_Unique ||
1226 : 1345 : indexInfo->ii_ExclusionOps != NULL);
1227 : :
1228 : : /*
1229 : : * "Any visible" mode is not compatible with uniqueness checks; make sure
1230 : : * only one of those is requested.
1231 : : */
1232 [ + + + - ]: 4057 : Assert(!(anyvisible && checking_uniqueness));
1233 : :
1234 : : /*
1235 : : * Need an EState for evaluation of index expressions and partial-index
1236 : : * predicates. Also a slot to hold the current tuple.
1237 : : */
1238 : 4057 : estate = CreateExecutorState();
1239 [ - + ]: 4057 : econtext = GetPerTupleExprContext(estate);
1240 : 4057 : slot = table_slot_create(heapRelation, NULL);
1241 : :
1242 : : /* Arrange for econtext's scan tuple to be the tuple under test */
1243 : 4057 : econtext->ecxt_scantuple = slot;
1244 : :
1245 : : /* Set up execution state for predicate, if any. */
1246 : 4057 : predicate = ExecPrepareQual(indexInfo->ii_Predicate, estate);
1247 : :
1248 : : /*
1249 : : * Prepare for scan of the base relation. In a normal index build, we use
1250 : : * SnapshotAny because we must retrieve all tuples and do our own time
1251 : : * qual checks (because we have to index RECENTLY_DEAD tuples). In a
1252 : : * concurrent build, or during bootstrap, we take a regular MVCC snapshot
1253 : : * and index whatever's live according to that.
1254 : : */
1255 : 4057 : OldestXmin = InvalidTransactionId;
1256 : :
1257 : : /* okay to ignore lazy VACUUMs here */
1258 [ + + + + ]: 4057 : if (!IsBootstrapProcessingMode() && !indexInfo->ii_Concurrent)
1259 : 3835 : OldestXmin = GetOldestNonRemovableTransactionId(heapRelation);
1260 : :
1261 [ + + ]: 4057 : if (!scan)
1262 : : {
1263 : : /*
1264 : : * Serial index build.
1265 : : *
1266 : : * Must begin our own heap scan in this case. We may also need to
1267 : : * register a snapshot whose lifetime is under our direct control.
1268 : : */
1269 [ + + ]: 4001 : if (!TransactionIdIsValid(OldestXmin))
1270 : : {
1271 : 222 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
1272 : 222 : need_unregister_snapshot = true;
1273 : 222 : }
1274 : : else
1275 : 3779 : snapshot = SnapshotAny;
1276 : :
1277 : 8002 : scan = table_beginscan_strat(heapRelation, /* relation */
1278 : 4001 : snapshot, /* snapshot */
1279 : : 0, /* number of keys */
1280 : : NULL, /* scan key */
1281 : : true, /* buffer access strategy OK */
1282 : 4001 : allow_sync); /* syncscan OK? */
1283 : 4001 : }
1284 : : else
1285 : : {
1286 : : /*
1287 : : * Parallel index build.
1288 : : *
1289 : : * Parallel case never registers/unregisters own snapshot. Snapshot
1290 : : * is taken from parallel heap scan, and is SnapshotAny or an MVCC
1291 : : * snapshot, based on same criteria as serial case.
1292 : : */
1293 [ + - ]: 56 : Assert(!IsBootstrapProcessingMode());
1294 [ + - ]: 56 : Assert(allow_sync);
1295 : 56 : snapshot = scan->rs_snapshot;
1296 : : }
1297 : :
1298 : 4057 : hscan = (HeapScanDesc) scan;
1299 : :
1300 : : /*
1301 : : * Must have called GetOldestNonRemovableTransactionId() if using
1302 : : * SnapshotAny. Shouldn't have for an MVCC snapshot. (It's especially
1303 : : * worth checking this for parallel builds, since ambuild routines that
1304 : : * support parallel builds must work these details out for themselves.)
1305 : : */
1306 [ + + - + : 4057 : Assert(snapshot == SnapshotAny || IsMVCCSnapshot(snapshot));
# # ]
1307 [ + + + - ]: 4057 : Assert(snapshot == SnapshotAny ? TransactionIdIsValid(OldestXmin) :
1308 : : !TransactionIdIsValid(OldestXmin));
1309 [ + + + - ]: 4057 : Assert(snapshot == SnapshotAny || !anyvisible);
1310 : :
1311 : : /* Publish number of blocks to scan */
1312 [ + + ]: 4057 : if (progress)
1313 : : {
1314 : 4010 : BlockNumber nblocks;
1315 : :
1316 [ + + ]: 4010 : if (hscan->rs_base.rs_parallel != NULL)
1317 : : {
1318 : 29 : ParallelBlockTableScanDesc pbscan;
1319 : :
1320 : 29 : pbscan = (ParallelBlockTableScanDesc) hscan->rs_base.rs_parallel;
1321 : 29 : nblocks = pbscan->phs_nblocks;
1322 : 29 : }
1323 : : else
1324 : 3981 : nblocks = hscan->rs_nblocks;
1325 : :
1326 : 4010 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_TOTAL,
1327 : 4010 : nblocks);
1328 : 4010 : }
1329 : :
1330 : : /* set our scan endpoints */
1331 [ + + ]: 4057 : if (!allow_sync)
1332 : 76 : heap_setscanlimits(scan, start_blockno, numblocks);
1333 : : else
1334 : : {
1335 : : /* syncscan can only be requested on whole relation */
1336 [ + - ]: 3981 : Assert(start_blockno == 0);
1337 [ + - ]: 3981 : Assert(numblocks == InvalidBlockNumber);
1338 : : }
1339 : :
1340 : 4057 : reltuples = 0;
1341 : :
1342 : : /*
1343 : : * Scan all tuples in the base relation.
1344 : : */
1345 [ + + ]: 1652759 : while ((heapTuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
1346 : : {
1347 : 1648702 : bool tupleIsAlive;
1348 : :
1349 [ - + ]: 1648702 : CHECK_FOR_INTERRUPTS();
1350 : :
1351 : : /* Report scan progress, if asked to. */
1352 [ + + ]: 1648702 : if (progress)
1353 : : {
1354 : 1569920 : BlockNumber blocks_done = heapam_scan_get_blocks_done(hscan);
1355 : :
1356 [ + + ]: 1569920 : if (blocks_done != previous_blkno)
1357 : : {
1358 : 18624 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_DONE,
1359 : 18624 : blocks_done);
1360 : 18624 : previous_blkno = blocks_done;
1361 : 18624 : }
1362 : 1569920 : }
1363 : :
1364 : : /*
1365 : : * When dealing with a HOT-chain of updated tuples, we want to index
1366 : : * the values of the live tuple (if any), but index it under the TID
1367 : : * of the chain's root tuple. This approach is necessary to preserve
1368 : : * the HOT-chain structure in the heap. So we need to be able to find
1369 : : * the root item offset for every tuple that's in a HOT-chain. When
1370 : : * first reaching a new page of the relation, call
1371 : : * heap_get_root_tuples() to build a map of root item offsets on the
1372 : : * page.
1373 : : *
1374 : : * It might look unsafe to use this information across buffer
1375 : : * lock/unlock. However, we hold ShareLock on the table so no
1376 : : * ordinary insert/update/delete should occur; and we hold pin on the
1377 : : * buffer continuously while visiting the page, so no pruning
1378 : : * operation can occur either.
1379 : : *
1380 : : * In cases with only ShareUpdateExclusiveLock on the table, it's
1381 : : * possible for some HOT tuples to appear that we didn't know about
1382 : : * when we first read the page. To handle that case, we re-obtain the
1383 : : * list of root offsets when a HOT tuple points to a root item that we
1384 : : * don't know about.
1385 : : *
1386 : : * Also, although our opinions about tuple liveness could change while
1387 : : * we scan the page (due to concurrent transaction commits/aborts),
1388 : : * the chain root locations won't, so this info doesn't need to be
1389 : : * rebuilt after waiting for another transaction.
1390 : : *
1391 : : * Note the implied assumption that there is no more than one live
1392 : : * tuple per HOT-chain --- else we could create more than one index
1393 : : * entry pointing to the same root tuple.
1394 : : */
1395 [ + + ]: 1648702 : if (hscan->rs_cblock != root_blkno)
1396 : : {
1397 : 19849 : Page page = BufferGetPage(hscan->rs_cbuf);
1398 : :
1399 : 19849 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
1400 : 19849 : heap_get_root_tuples(page, root_offsets);
1401 : 19849 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1402 : :
1403 : 19849 : root_blkno = hscan->rs_cblock;
1404 : 19849 : }
1405 : :
1406 [ + + ]: 1648702 : if (snapshot == SnapshotAny)
1407 : : {
1408 : : /* do our own time qual check */
1409 : 1622818 : bool indexIt;
1410 : 1622818 : TransactionId xwait;
1411 : :
1412 : : recheck:
1413 : :
1414 : : /*
1415 : : * We could possibly get away with not locking the buffer here,
1416 : : * since caller should hold ShareLock on the relation, but let's
1417 : : * be conservative about it. (This remark is still correct even
1418 : : * with HOT-pruning: our pin on the buffer prevents pruning.)
1419 : : */
1420 : 1622818 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
1421 : :
1422 : : /*
1423 : : * The criteria for counting a tuple as live in this block need to
1424 : : * match what analyze.c's heapam_scan_analyze_next_tuple() does,
1425 : : * otherwise CREATE INDEX and ANALYZE may produce wildly different
1426 : : * reltuples values, e.g. when there are many recently-dead
1427 : : * tuples.
1428 : : */
1429 [ + + + + : 3245636 : switch (HeapTupleSatisfiesVacuum(heapTuple, OldestXmin,
+ - + + +
+ + - ]
1430 : 1622818 : hscan->rs_cbuf))
1431 : : {
1432 : : case HEAPTUPLE_DEAD:
1433 : : /* Definitely dead, we can ignore it */
1434 : 85 : indexIt = false;
1435 : 85 : tupleIsAlive = false;
1436 : 85 : break;
1437 : : case HEAPTUPLE_LIVE:
1438 : : /* Normal case, index and unique-check it */
1439 : 1471436 : indexIt = true;
1440 : 1471436 : tupleIsAlive = true;
1441 : : /* Count it as live, too */
1442 : 1471436 : reltuples += 1;
1443 : 1471436 : break;
1444 : : case HEAPTUPLE_RECENTLY_DEAD:
1445 : :
1446 : : /*
1447 : : * If tuple is recently deleted then we must index it
1448 : : * anyway to preserve MVCC semantics. (Pre-existing
1449 : : * transactions could try to use the index after we finish
1450 : : * building it, and may need to see such tuples.)
1451 : : *
1452 : : * However, if it was HOT-updated then we must only index
1453 : : * the live tuple at the end of the HOT-chain. Since this
1454 : : * breaks semantics for pre-existing snapshots, mark the
1455 : : * index as unusable for them.
1456 : : *
1457 : : * We don't count recently-dead tuples in reltuples, even
1458 : : * if we index them; see heapam_scan_analyze_next_tuple().
1459 : : */
1460 [ + + ]: 38612 : if (HeapTupleIsHotUpdated(heapTuple))
1461 : : {
1462 : 33 : indexIt = false;
1463 : : /* mark the index as unsafe for old snapshots */
1464 : 33 : indexInfo->ii_BrokenHotChain = true;
1465 : 33 : }
1466 : : else
1467 : 38579 : indexIt = true;
1468 : : /* In any case, exclude the tuple from unique-checking */
1469 : 38612 : tupleIsAlive = false;
1470 : 38612 : break;
1471 : : case HEAPTUPLE_INSERT_IN_PROGRESS:
1472 : :
1473 : : /*
1474 : : * In "anyvisible" mode, this tuple is visible and we
1475 : : * don't need any further checks.
1476 : : */
1477 [ + - ]: 112680 : if (anyvisible)
1478 : : {
1479 : 0 : indexIt = true;
1480 : 0 : tupleIsAlive = true;
1481 : 0 : reltuples += 1;
1482 : 0 : break;
1483 : : }
1484 : :
1485 : : /*
1486 : : * Since caller should hold ShareLock or better, normally
1487 : : * the only way to see this is if it was inserted earlier
1488 : : * in our own transaction. However, it can happen in
1489 : : * system catalogs, since we tend to release write lock
1490 : : * before commit there. Give a warning if neither case
1491 : : * applies.
1492 : : */
1493 : 112680 : xwait = HeapTupleHeaderGetXmin(heapTuple->t_data);
1494 [ - + ]: 112680 : if (!TransactionIdIsCurrentTransactionId(xwait))
1495 : : {
1496 [ # # ]: 0 : if (!is_system_catalog)
1497 [ # # # # ]: 0 : elog(WARNING, "concurrent insert in progress within table \"%s\"",
1498 : : RelationGetRelationName(heapRelation));
1499 : :
1500 : : /*
1501 : : * If we are performing uniqueness checks, indexing
1502 : : * such a tuple could lead to a bogus uniqueness
1503 : : * failure. In that case we wait for the inserting
1504 : : * transaction to finish and check again.
1505 : : */
1506 [ # # ]: 0 : if (checking_uniqueness)
1507 : : {
1508 : : /*
1509 : : * Must drop the lock on the buffer before we wait
1510 : : */
1511 : 0 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1512 : 0 : XactLockTableWait(xwait, heapRelation,
1513 : 0 : &heapTuple->t_self,
1514 : : XLTW_InsertIndexUnique);
1515 [ # # ]: 0 : CHECK_FOR_INTERRUPTS();
1516 : 0 : goto recheck;
1517 : : }
1518 : 0 : }
1519 : : else
1520 : : {
1521 : : /*
1522 : : * For consistency with
1523 : : * heapam_scan_analyze_next_tuple(), count
1524 : : * HEAPTUPLE_INSERT_IN_PROGRESS tuples as live only
1525 : : * when inserted by our own transaction.
1526 : : */
1527 : 112680 : reltuples += 1;
1528 : : }
1529 : :
1530 : : /*
1531 : : * We must index such tuples, since if the index build
1532 : : * commits then they're good.
1533 : : */
1534 : 112680 : indexIt = true;
1535 : 112680 : tupleIsAlive = true;
1536 : 112680 : break;
1537 : : case HEAPTUPLE_DELETE_IN_PROGRESS:
1538 : :
1539 : : /*
1540 : : * As with INSERT_IN_PROGRESS case, this is unexpected
1541 : : * unless it's our own deletion or a system catalog; but
1542 : : * in anyvisible mode, this tuple is visible.
1543 : : */
1544 [ + - ]: 5 : if (anyvisible)
1545 : : {
1546 : 0 : indexIt = true;
1547 : 0 : tupleIsAlive = false;
1548 : 0 : reltuples += 1;
1549 : 0 : break;
1550 : : }
1551 : :
1552 : 5 : xwait = HeapTupleHeaderGetUpdateXid(heapTuple->t_data);
1553 [ - + ]: 5 : if (!TransactionIdIsCurrentTransactionId(xwait))
1554 : : {
1555 [ # # ]: 0 : if (!is_system_catalog)
1556 [ # # # # ]: 0 : elog(WARNING, "concurrent delete in progress within table \"%s\"",
1557 : : RelationGetRelationName(heapRelation));
1558 : :
1559 : : /*
1560 : : * If we are performing uniqueness checks, assuming
1561 : : * the tuple is dead could lead to missing a
1562 : : * uniqueness violation. In that case we wait for the
1563 : : * deleting transaction to finish and check again.
1564 : : *
1565 : : * Also, if it's a HOT-updated tuple, we should not
1566 : : * index it but rather the live tuple at the end of
1567 : : * the HOT-chain. However, the deleting transaction
1568 : : * could abort, possibly leaving this tuple as live
1569 : : * after all, in which case it has to be indexed. The
1570 : : * only way to know what to do is to wait for the
1571 : : * deleting transaction to finish and check again.
1572 : : */
1573 [ # # # # ]: 0 : if (checking_uniqueness ||
1574 : 0 : HeapTupleIsHotUpdated(heapTuple))
1575 : : {
1576 : : /*
1577 : : * Must drop the lock on the buffer before we wait
1578 : : */
1579 : 0 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1580 : 0 : XactLockTableWait(xwait, heapRelation,
1581 : 0 : &heapTuple->t_self,
1582 : : XLTW_InsertIndexUnique);
1583 [ # # ]: 0 : CHECK_FOR_INTERRUPTS();
1584 : 0 : goto recheck;
1585 : : }
1586 : :
1587 : : /*
1588 : : * Otherwise index it but don't check for uniqueness,
1589 : : * the same as a RECENTLY_DEAD tuple.
1590 : : */
1591 : 0 : indexIt = true;
1592 : :
1593 : : /*
1594 : : * Count HEAPTUPLE_DELETE_IN_PROGRESS tuples as live,
1595 : : * if they were not deleted by the current
1596 : : * transaction. That's what
1597 : : * heapam_scan_analyze_next_tuple() does, and we want
1598 : : * the behavior to be consistent.
1599 : : */
1600 : 0 : reltuples += 1;
1601 : 0 : }
1602 [ - + ]: 5 : else if (HeapTupleIsHotUpdated(heapTuple))
1603 : : {
1604 : : /*
1605 : : * It's a HOT-updated tuple deleted by our own xact.
1606 : : * We can assume the deletion will commit (else the
1607 : : * index contents don't matter), so treat the same as
1608 : : * RECENTLY_DEAD HOT-updated tuples.
1609 : : */
1610 : 0 : indexIt = false;
1611 : : /* mark the index as unsafe for old snapshots */
1612 : 0 : indexInfo->ii_BrokenHotChain = true;
1613 : 0 : }
1614 : : else
1615 : : {
1616 : : /*
1617 : : * It's a regular tuple deleted by our own xact. Index
1618 : : * it, but don't check for uniqueness nor count in
1619 : : * reltuples, the same as a RECENTLY_DEAD tuple.
1620 : : */
1621 : 5 : indexIt = true;
1622 : : }
1623 : : /* In any case, exclude the tuple from unique-checking */
1624 : 5 : tupleIsAlive = false;
1625 : 5 : break;
1626 : : default:
1627 [ # # # # ]: 0 : elog(ERROR, "unexpected HeapTupleSatisfiesVacuum result");
1628 : 0 : indexIt = tupleIsAlive = false; /* keep compiler quiet */
1629 : 0 : break;
1630 : : }
1631 : :
1632 : 1622818 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1633 : :
1634 [ + + ]: 1622818 : if (!indexIt)
1635 : 118 : continue;
1636 [ + + ]: 1622818 : }
1637 : : else
1638 : : {
1639 : : /* heap_getnext did the time qual check */
1640 : 25884 : tupleIsAlive = true;
1641 : 25884 : reltuples += 1;
1642 : : }
1643 : :
1644 : 1648584 : MemoryContextReset(econtext->ecxt_per_tuple_memory);
1645 : :
1646 : : /* Set up for predicate or expression evaluation */
1647 : 1648584 : ExecStoreBufferHeapTuple(heapTuple, slot, hscan->rs_cbuf);
1648 : :
1649 : : /*
1650 : : * In a partial index, discard tuples that don't satisfy the
1651 : : * predicate.
1652 : : */
1653 [ + + ]: 1648584 : if (predicate != NULL)
1654 : : {
1655 [ + + ]: 28091 : if (!ExecQual(predicate, econtext))
1656 : 16924 : continue;
1657 : 11167 : }
1658 : :
1659 : : /*
1660 : : * For the current heap tuple, extract all the attributes we use in
1661 : : * this index, and note which are null. This also performs evaluation
1662 : : * of any expressions needed.
1663 : : */
1664 : 3263320 : FormIndexDatum(indexInfo,
1665 : 1631660 : slot,
1666 : 1631660 : estate,
1667 : 1631660 : values,
1668 : 1631660 : isnull);
1669 : :
1670 : : /*
1671 : : * You'd think we should go ahead and build the index tuple here, but
1672 : : * some index AMs want to do further processing on the data first. So
1673 : : * pass the values[] and isnull[] arrays, instead.
1674 : : */
1675 : :
1676 [ + + ]: 1631660 : if (HeapTupleIsHeapOnly(heapTuple))
1677 : : {
1678 : : /*
1679 : : * For a heap-only tuple, pretend its TID is that of the root. See
1680 : : * src/backend/access/heap/README.HOT for discussion.
1681 : : */
1682 : 47 : ItemPointerData tid;
1683 : 47 : OffsetNumber offnum;
1684 : :
1685 : 47 : offnum = ItemPointerGetOffsetNumber(&heapTuple->t_self);
1686 : :
1687 : : /*
1688 : : * If a HOT tuple points to a root that we don't know about,
1689 : : * obtain root items afresh. If that still fails, report it as
1690 : : * corruption.
1691 : : */
1692 [ + - ]: 47 : if (root_offsets[offnum - 1] == InvalidOffsetNumber)
1693 : : {
1694 : 0 : Page page = BufferGetPage(hscan->rs_cbuf);
1695 : :
1696 : 0 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
1697 : 0 : heap_get_root_tuples(page, root_offsets);
1698 : 0 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1699 : 0 : }
1700 : :
1701 [ - + + - ]: 47 : if (!OffsetNumberIsValid(root_offsets[offnum - 1]))
1702 [ # # # # ]: 0 : ereport(ERROR,
1703 : : (errcode(ERRCODE_DATA_CORRUPTED),
1704 : : errmsg_internal("failed to find parent tuple for heap-only tuple at (%u,%u) in table \"%s\"",
1705 : : ItemPointerGetBlockNumber(&heapTuple->t_self),
1706 : : offnum,
1707 : : RelationGetRelationName(heapRelation))));
1708 : :
1709 : 94 : ItemPointerSet(&tid, ItemPointerGetBlockNumber(&heapTuple->t_self),
1710 : 47 : root_offsets[offnum - 1]);
1711 : :
1712 : : /* Call the AM's callback routine to process the tuple */
1713 : 94 : callback(indexRelation, &tid, values, isnull, tupleIsAlive,
1714 : 47 : callback_state);
1715 : 47 : }
1716 : : else
1717 : : {
1718 : : /* Call the AM's callback routine to process the tuple */
1719 : 3263226 : callback(indexRelation, &heapTuple->t_self, values, isnull,
1720 : 1631613 : tupleIsAlive, callback_state);
1721 : : }
1722 [ - + + ]: 1648702 : }
1723 : :
1724 : : /* Report scan progress one last time. */
1725 [ + + ]: 4057 : if (progress)
1726 : : {
1727 : 4008 : BlockNumber blks_done;
1728 : :
1729 [ + + ]: 4008 : if (hscan->rs_base.rs_parallel != NULL)
1730 : : {
1731 : 29 : ParallelBlockTableScanDesc pbscan;
1732 : :
1733 : 29 : pbscan = (ParallelBlockTableScanDesc) hscan->rs_base.rs_parallel;
1734 : 29 : blks_done = pbscan->phs_nblocks;
1735 : 29 : }
1736 : : else
1737 : 3979 : blks_done = hscan->rs_nblocks;
1738 : :
1739 : 4008 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_DONE,
1740 : 4008 : blks_done);
1741 : 4008 : }
1742 : :
1743 : 4057 : table_endscan(scan);
1744 : :
1745 : : /* we can now forget our snapshot, if set and registered by us */
1746 [ + + ]: 4057 : if (need_unregister_snapshot)
1747 : 221 : UnregisterSnapshot(snapshot);
1748 : :
1749 : 4057 : ExecDropSingleTupleTableSlot(slot);
1750 : :
1751 : 4057 : FreeExecutorState(estate);
1752 : :
1753 : : /* These may have been pointing to the now-gone estate */
1754 : 4057 : indexInfo->ii_ExpressionsState = NIL;
1755 : 4057 : indexInfo->ii_PredicateState = NULL;
1756 : :
1757 : 8114 : return reltuples;
1758 : 4057 : }
1759 : :
1760 : : static void
1761 : 59 : heapam_index_validate_scan(Relation heapRelation,
1762 : : Relation indexRelation,
1763 : : IndexInfo *indexInfo,
1764 : : Snapshot snapshot,
1765 : : ValidateIndexState *state)
1766 : : {
1767 : 59 : TableScanDesc scan;
1768 : 59 : HeapScanDesc hscan;
1769 : 59 : HeapTuple heapTuple;
1770 : 59 : Datum values[INDEX_MAX_KEYS];
1771 : 59 : bool isnull[INDEX_MAX_KEYS];
1772 : 59 : ExprState *predicate;
1773 : 59 : TupleTableSlot *slot;
1774 : 59 : EState *estate;
1775 : 59 : ExprContext *econtext;
1776 : 59 : BlockNumber root_blkno = InvalidBlockNumber;
1777 : 59 : OffsetNumber root_offsets[MaxHeapTuplesPerPage];
1778 : 59 : bool in_index[MaxHeapTuplesPerPage];
1779 : 59 : BlockNumber previous_blkno = InvalidBlockNumber;
1780 : :
1781 : : /* state variables for the merge */
1782 : 59 : ItemPointer indexcursor = NULL;
1783 : 59 : ItemPointerData decoded;
1784 : 59 : bool tuplesort_empty = false;
1785 : :
1786 : : /*
1787 : : * sanity checks
1788 : : */
1789 [ + - ]: 59 : Assert(OidIsValid(indexRelation->rd_rel->relam));
1790 : :
1791 : : /*
1792 : : * Need an EState for evaluation of index expressions and partial-index
1793 : : * predicates. Also a slot to hold the current tuple.
1794 : : */
1795 : 59 : estate = CreateExecutorState();
1796 [ - + ]: 59 : econtext = GetPerTupleExprContext(estate);
1797 : 59 : slot = MakeSingleTupleTableSlot(RelationGetDescr(heapRelation),
1798 : : &TTSOpsHeapTuple);
1799 : :
1800 : : /* Arrange for econtext's scan tuple to be the tuple under test */
1801 : 59 : econtext->ecxt_scantuple = slot;
1802 : :
1803 : : /* Set up execution state for predicate, if any. */
1804 : 59 : predicate = ExecPrepareQual(indexInfo->ii_Predicate, estate);
1805 : :
1806 : : /*
1807 : : * Prepare for scan of the base relation. We need just those tuples
1808 : : * satisfying the passed-in reference snapshot. We must disable syncscan
1809 : : * here, because it's critical that we read from block zero forward to
1810 : : * match the sorted TIDs.
1811 : : */
1812 : 118 : scan = table_beginscan_strat(heapRelation, /* relation */
1813 : 59 : snapshot, /* snapshot */
1814 : : 0, /* number of keys */
1815 : : NULL, /* scan key */
1816 : : true, /* buffer access strategy OK */
1817 : : false); /* syncscan not OK */
1818 : 59 : hscan = (HeapScanDesc) scan;
1819 : :
1820 : 59 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_TOTAL,
1821 : 59 : hscan->rs_nblocks);
1822 : :
1823 : : /*
1824 : : * Scan all tuples matching the snapshot.
1825 : : */
1826 [ + + ]: 2730 : while ((heapTuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
1827 : : {
1828 : 2671 : ItemPointer heapcursor = &heapTuple->t_self;
1829 : 2671 : ItemPointerData rootTuple;
1830 : 2671 : OffsetNumber root_offnum;
1831 : :
1832 [ + - ]: 2671 : CHECK_FOR_INTERRUPTS();
1833 : :
1834 : 2671 : state->htups += 1;
1835 : :
1836 [ + + + + ]: 2671 : if ((previous_blkno == InvalidBlockNumber) ||
1837 : 2644 : (hscan->rs_cblock != previous_blkno))
1838 : : {
1839 : 40 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_DONE,
1840 : 40 : hscan->rs_cblock);
1841 : 40 : previous_blkno = hscan->rs_cblock;
1842 : 40 : }
1843 : :
1844 : : /*
1845 : : * As commented in table_index_build_scan, we should index heap-only
1846 : : * tuples under the TIDs of their root tuples; so when we advance onto
1847 : : * a new heap page, build a map of root item offsets on the page.
1848 : : *
1849 : : * This complicates merging against the tuplesort output: we will
1850 : : * visit the live tuples in order by their offsets, but the root
1851 : : * offsets that we need to compare against the index contents might be
1852 : : * ordered differently. So we might have to "look back" within the
1853 : : * tuplesort output, but only within the current page. We handle that
1854 : : * by keeping a bool array in_index[] showing all the
1855 : : * already-passed-over tuplesort output TIDs of the current page. We
1856 : : * clear that array here, when advancing onto a new heap page.
1857 : : */
1858 [ + + ]: 2671 : if (hscan->rs_cblock != root_blkno)
1859 : : {
1860 : 40 : Page page = BufferGetPage(hscan->rs_cbuf);
1861 : :
1862 : 40 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
1863 : 40 : heap_get_root_tuples(page, root_offsets);
1864 : 40 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1865 : :
1866 : 40 : memset(in_index, 0, sizeof(in_index));
1867 : :
1868 : 40 : root_blkno = hscan->rs_cblock;
1869 : 40 : }
1870 : :
1871 : : /* Convert actual tuple TID to root TID */
1872 : 2671 : rootTuple = *heapcursor;
1873 : 2671 : root_offnum = ItemPointerGetOffsetNumber(heapcursor);
1874 : :
1875 [ + + ]: 2671 : if (HeapTupleIsHeapOnly(heapTuple))
1876 : : {
1877 : 1 : root_offnum = root_offsets[root_offnum - 1];
1878 [ - + + - ]: 1 : if (!OffsetNumberIsValid(root_offnum))
1879 [ # # # # ]: 0 : ereport(ERROR,
1880 : : (errcode(ERRCODE_DATA_CORRUPTED),
1881 : : errmsg_internal("failed to find parent tuple for heap-only tuple at (%u,%u) in table \"%s\"",
1882 : : ItemPointerGetBlockNumber(heapcursor),
1883 : : ItemPointerGetOffsetNumber(heapcursor),
1884 : : RelationGetRelationName(heapRelation))));
1885 : 1 : ItemPointerSetOffsetNumber(&rootTuple, root_offnum);
1886 : 1 : }
1887 : :
1888 : : /*
1889 : : * "merge" by skipping through the index tuples until we find or pass
1890 : : * the current root tuple.
1891 : : */
1892 [ + + + + ]: 10667 : while (!tuplesort_empty &&
1893 [ + + ]: 5330 : (!indexcursor ||
1894 : 5303 : ItemPointerCompare(indexcursor, &rootTuple) < 0))
1895 : : {
1896 : 2666 : Datum ts_val;
1897 : 2666 : bool ts_isnull;
1898 : :
1899 [ + + ]: 2666 : if (indexcursor)
1900 : : {
1901 : : /*
1902 : : * Remember index items seen earlier on the current heap page
1903 : : */
1904 [ + + ]: 2639 : if (ItemPointerGetBlockNumber(indexcursor) == root_blkno)
1905 : 2626 : in_index[ItemPointerGetOffsetNumber(indexcursor) - 1] = true;
1906 : 2639 : }
1907 : :
1908 : 2666 : tuplesort_empty = !tuplesort_getdatum(state->tuplesort, true,
1909 : : false, &ts_val, &ts_isnull,
1910 : : NULL);
1911 [ + + - + ]: 2666 : Assert(tuplesort_empty || !ts_isnull);
1912 [ + + ]: 2666 : if (!tuplesort_empty)
1913 : : {
1914 : 2662 : itemptr_decode(&decoded, DatumGetInt64(ts_val));
1915 : 2662 : indexcursor = &decoded;
1916 : 2662 : }
1917 : : else
1918 : : {
1919 : : /* Be tidy */
1920 : 4 : indexcursor = NULL;
1921 : : }
1922 : 2666 : }
1923 : :
1924 : : /*
1925 : : * If the tuplesort has overshot *and* we didn't see a match earlier,
1926 : : * then this tuple is missing from the index, so insert it.
1927 : : */
1928 [ + + ]: 2671 : if ((tuplesort_empty ||
1929 [ + + ]: 2671 : ItemPointerCompare(indexcursor, &rootTuple) > 0) &&
1930 : 2671 : !in_index[root_offnum - 1])
1931 : : {
1932 : 9 : MemoryContextReset(econtext->ecxt_per_tuple_memory);
1933 : :
1934 : : /* Set up for predicate or expression evaluation */
1935 : 9 : ExecStoreHeapTuple(heapTuple, slot, false);
1936 : :
1937 : : /*
1938 : : * In a partial index, discard tuples that don't satisfy the
1939 : : * predicate.
1940 : : */
1941 [ + + ]: 9 : if (predicate != NULL)
1942 : : {
1943 [ - + ]: 8 : if (!ExecQual(predicate, econtext))
1944 : 8 : continue;
1945 : 0 : }
1946 : :
1947 : : /*
1948 : : * For the current heap tuple, extract all the attributes we use
1949 : : * in this index, and note which are null. This also performs
1950 : : * evaluation of any expressions needed.
1951 : : */
1952 : 2 : FormIndexDatum(indexInfo,
1953 : 1 : slot,
1954 : 1 : estate,
1955 : 1 : values,
1956 : 1 : isnull);
1957 : :
1958 : : /*
1959 : : * You'd think we should go ahead and build the index tuple here,
1960 : : * but some index AMs want to do further processing on the data
1961 : : * first. So pass the values[] and isnull[] arrays, instead.
1962 : : */
1963 : :
1964 : : /*
1965 : : * If the tuple is already committed dead, you might think we
1966 : : * could suppress uniqueness checking, but this is no longer true
1967 : : * in the presence of HOT, because the insert is actually a proxy
1968 : : * for a uniqueness check on the whole HOT-chain. That is, the
1969 : : * tuple we have here could be dead because it was already
1970 : : * HOT-updated, and if so the updating transaction will not have
1971 : : * thought it should insert index entries. The index AM will
1972 : : * check the whole HOT-chain and correctly detect a conflict if
1973 : : * there is one.
1974 : : */
1975 : :
1976 : 2 : index_insert(indexRelation,
1977 : 1 : values,
1978 : 1 : isnull,
1979 : : &rootTuple,
1980 : 1 : heapRelation,
1981 : 1 : indexInfo->ii_Unique ?
1982 : : UNIQUE_CHECK_YES : UNIQUE_CHECK_NO,
1983 : : false,
1984 : 1 : indexInfo);
1985 : :
1986 : 1 : state->tups_inserted += 1;
1987 : 1 : }
1988 [ - + + ]: 2671 : }
1989 : :
1990 : 59 : table_endscan(scan);
1991 : :
1992 : 59 : ExecDropSingleTupleTableSlot(slot);
1993 : :
1994 : 59 : FreeExecutorState(estate);
1995 : :
1996 : : /* These may have been pointing to the now-gone estate */
1997 : 59 : indexInfo->ii_ExpressionsState = NIL;
1998 : 59 : indexInfo->ii_PredicateState = NULL;
1999 : 59 : }
2000 : :
2001 : : /*
2002 : : * Return the number of blocks that have been read by this scan since
2003 : : * starting. This is meant for progress reporting rather than be fully
2004 : : * accurate: in a parallel scan, workers can be concurrently reading blocks
2005 : : * further ahead than what we report.
2006 : : */
2007 : : static BlockNumber
2008 : 1569920 : heapam_scan_get_blocks_done(HeapScanDesc hscan)
2009 : : {
2010 : 1569920 : ParallelBlockTableScanDesc bpscan = NULL;
2011 : 1569920 : BlockNumber startblock;
2012 : 1569920 : BlockNumber blocks_done;
2013 : :
2014 [ + + ]: 1569920 : if (hscan->rs_base.rs_parallel != NULL)
2015 : : {
2016 : 171262 : bpscan = (ParallelBlockTableScanDesc) hscan->rs_base.rs_parallel;
2017 : 171262 : startblock = bpscan->phs_startblock;
2018 : 171262 : }
2019 : : else
2020 : 1398658 : startblock = hscan->rs_startblock;
2021 : :
2022 : : /*
2023 : : * Might have wrapped around the end of the relation, if startblock was
2024 : : * not zero.
2025 : : */
2026 [ + + ]: 1569920 : if (hscan->rs_cblock > startblock)
2027 : 1536519 : blocks_done = hscan->rs_cblock - startblock;
2028 : : else
2029 : : {
2030 : 33401 : BlockNumber nblocks;
2031 : :
2032 [ + + ]: 33401 : nblocks = bpscan != NULL ? bpscan->phs_nblocks : hscan->rs_nblocks;
2033 : 66802 : blocks_done = nblocks - startblock +
2034 : 33401 : hscan->rs_cblock;
2035 : 33401 : }
2036 : :
2037 : 3139840 : return blocks_done;
2038 : 1569920 : }
2039 : :
2040 : :
2041 : : /* ------------------------------------------------------------------------
2042 : : * Miscellaneous callbacks for the heap AM
2043 : : * ------------------------------------------------------------------------
2044 : : */
2045 : :
2046 : : /*
2047 : : * Check to see whether the table needs a TOAST table. It does only if
2048 : : * (1) there are any toastable attributes, and (2) the maximum length
2049 : : * of a tuple could exceed TOAST_TUPLE_THRESHOLD. (We don't want to
2050 : : * create a toast table for something like "f1 varchar(20)".)
2051 : : */
2052 : : static bool
2053 : 4953 : heapam_relation_needs_toast_table(Relation rel)
2054 : : {
2055 : 4953 : int32 data_length = 0;
2056 : 4953 : bool maxlength_unknown = false;
2057 : 4953 : bool has_toastable_attrs = false;
2058 : 4953 : TupleDesc tupdesc = rel->rd_att;
2059 : 4953 : int32 tuple_length;
2060 : 4953 : int i;
2061 : :
2062 [ + + ]: 17390 : for (i = 0; i < tupdesc->natts; i++)
2063 : : {
2064 : 12437 : Form_pg_attribute att = TupleDescAttr(tupdesc, i);
2065 : :
2066 [ + + ]: 12437 : if (att->attisdropped)
2067 : 178 : continue;
2068 [ + + ]: 12259 : if (att->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL)
2069 : 139 : continue;
2070 [ + + + + : 12120 : data_length = att_align_nominal(data_length, att->attalign);
+ + - + ]
2071 [ + + ]: 12120 : if (att->attlen > 0)
2072 : : {
2073 : : /* Fixed-length types are never toastable */
2074 : 9578 : data_length += att->attlen;
2075 : 9578 : }
2076 : : else
2077 : : {
2078 : 5084 : int32 maxlen = type_maximum_size(att->atttypid,
2079 : 2542 : att->atttypmod);
2080 : :
2081 [ + + ]: 2542 : if (maxlen < 0)
2082 : 2145 : maxlength_unknown = true;
2083 : : else
2084 : 397 : data_length += maxlen;
2085 [ + + ]: 2542 : if (att->attstorage != TYPSTORAGE_PLAIN)
2086 : 2521 : has_toastable_attrs = true;
2087 : 2542 : }
2088 [ - + + ]: 12437 : }
2089 [ + + ]: 4953 : if (!has_toastable_attrs)
2090 : 3172 : return false; /* nothing to toast? */
2091 [ + + ]: 1781 : if (maxlength_unknown)
2092 : 1497 : return true; /* any unlimited-length attrs? */
2093 : 568 : tuple_length = MAXALIGN(SizeofHeapTupleHeader +
2094 : 284 : BITMAPLEN(tupdesc->natts)) +
2095 : 284 : MAXALIGN(data_length);
2096 : 284 : return (tuple_length > TOAST_TUPLE_THRESHOLD);
2097 : 4953 : }
2098 : :
2099 : : /*
2100 : : * TOAST tables for heap relations are just heap relations.
2101 : : */
2102 : : static Oid
2103 : 1499 : heapam_relation_toast_am(Relation rel)
2104 : : {
2105 : 1499 : return rel->rd_rel->relam;
2106 : : }
2107 : :
2108 : :
2109 : : /* ------------------------------------------------------------------------
2110 : : * Planner related callbacks for the heap AM
2111 : : * ------------------------------------------------------------------------
2112 : : */
2113 : :
2114 : : #define HEAP_OVERHEAD_BYTES_PER_TUPLE \
2115 : : (MAXALIGN(SizeofHeapTupleHeader) + sizeof(ItemIdData))
2116 : : #define HEAP_USABLE_BYTES_PER_PAGE \
2117 : : (BLCKSZ - SizeOfPageHeaderData)
2118 : :
2119 : : static void
2120 : 47801 : heapam_estimate_rel_size(Relation rel, int32 *attr_widths,
2121 : : BlockNumber *pages, double *tuples,
2122 : : double *allvisfrac)
2123 : : {
2124 : 95602 : table_block_relation_estimate_size(rel, attr_widths, pages,
2125 : 47801 : tuples, allvisfrac,
2126 : : HEAP_OVERHEAD_BYTES_PER_TUPLE,
2127 : : HEAP_USABLE_BYTES_PER_PAGE);
2128 : 47801 : }
2129 : :
2130 : :
2131 : : /* ------------------------------------------------------------------------
2132 : : * Executor related callbacks for the heap AM
2133 : : * ------------------------------------------------------------------------
2134 : : */
2135 : :
2136 : : static bool
2137 : 678414 : heapam_scan_bitmap_next_tuple(TableScanDesc scan,
2138 : : TupleTableSlot *slot,
2139 : : bool *recheck,
2140 : : uint64 *lossy_pages,
2141 : : uint64 *exact_pages)
2142 : : {
2143 : 678414 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) scan;
2144 : 678414 : HeapScanDesc hscan = (HeapScanDesc) bscan;
2145 : 678414 : OffsetNumber targoffset;
2146 : 678414 : Page page;
2147 : 678414 : ItemId lp;
2148 : :
2149 : : /*
2150 : : * Out of range? If so, nothing more to look at on this page
2151 : : */
2152 [ + + ]: 720887 : while (hscan->rs_cindex >= hscan->rs_ntuples)
2153 : : {
2154 : : /*
2155 : : * Returns false if the bitmap is exhausted and there are no further
2156 : : * blocks we need to scan.
2157 : : */
2158 [ + + ]: 44425 : if (!BitmapHeapScanNextBlock(scan, recheck, lossy_pages, exact_pages))
2159 : 1952 : return false;
2160 : : }
2161 : :
2162 : 676462 : targoffset = hscan->rs_vistuples[hscan->rs_cindex];
2163 : 676462 : page = BufferGetPage(hscan->rs_cbuf);
2164 : 676462 : lp = PageGetItemId(page, targoffset);
2165 [ + - ]: 676462 : Assert(ItemIdIsNormal(lp));
2166 : :
2167 : 676462 : hscan->rs_ctup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2168 : 676462 : hscan->rs_ctup.t_len = ItemIdGetLength(lp);
2169 : 676462 : hscan->rs_ctup.t_tableOid = scan->rs_rd->rd_id;
2170 : 676462 : ItemPointerSet(&hscan->rs_ctup.t_self, hscan->rs_cblock, targoffset);
2171 : :
2172 [ + - + - : 676462 : pgstat_count_heap_fetch(scan->rs_rd);
# # ]
2173 : :
2174 : : /*
2175 : : * Set up the result slot to point to this tuple. Note that the slot
2176 : : * acquires a pin on the buffer.
2177 : : */
2178 : 1352924 : ExecStoreBufferHeapTuple(&hscan->rs_ctup,
2179 : 676462 : slot,
2180 : 676462 : hscan->rs_cbuf);
2181 : :
2182 : 676462 : hscan->rs_cindex++;
2183 : :
2184 : 676462 : return true;
2185 : 678414 : }
2186 : :
2187 : : static bool
2188 : 2132 : heapam_scan_sample_next_block(TableScanDesc scan, SampleScanState *scanstate)
2189 : : {
2190 : 2132 : HeapScanDesc hscan = (HeapScanDesc) scan;
2191 : 2132 : TsmRoutine *tsm = scanstate->tsmroutine;
2192 : 2132 : BlockNumber blockno;
2193 : :
2194 : : /* return false immediately if relation is empty */
2195 [ + - ]: 2132 : if (hscan->rs_nblocks == 0)
2196 : 0 : return false;
2197 : :
2198 : : /* release previous scan buffer, if any */
2199 [ + + ]: 2132 : if (BufferIsValid(hscan->rs_cbuf))
2200 : : {
2201 : 2107 : ReleaseBuffer(hscan->rs_cbuf);
2202 : 2107 : hscan->rs_cbuf = InvalidBuffer;
2203 : 2107 : }
2204 : :
2205 [ + + ]: 2132 : if (tsm->NextSampleBlock)
2206 : 721 : blockno = tsm->NextSampleBlock(scanstate, hscan->rs_nblocks);
2207 : : else
2208 : : {
2209 : : /* scanning table sequentially */
2210 : :
2211 [ + + ]: 1411 : if (hscan->rs_cblock == InvalidBlockNumber)
2212 : : {
2213 [ + - ]: 13 : Assert(!hscan->rs_inited);
2214 : 13 : blockno = hscan->rs_startblock;
2215 : 13 : }
2216 : : else
2217 : : {
2218 [ + - ]: 1398 : Assert(hscan->rs_inited);
2219 : :
2220 : 1398 : blockno = hscan->rs_cblock + 1;
2221 : :
2222 [ + + ]: 1398 : if (blockno >= hscan->rs_nblocks)
2223 : : {
2224 : : /* wrap to beginning of rel, might not have started at 0 */
2225 : 13 : blockno = 0;
2226 : 13 : }
2227 : :
2228 : : /*
2229 : : * Report our new scan position for synchronization purposes.
2230 : : *
2231 : : * Note: we do this before checking for end of scan so that the
2232 : : * final state of the position hint is back at the start of the
2233 : : * rel. That's not strictly necessary, but otherwise when you run
2234 : : * the same query multiple times the starting position would shift
2235 : : * a little bit backwards on every invocation, which is confusing.
2236 : : * We don't guarantee any specific ordering in general, though.
2237 : : */
2238 [ + - ]: 1398 : if (scan->rs_flags & SO_ALLOW_SYNC)
2239 : 0 : ss_report_location(scan->rs_rd, blockno);
2240 : :
2241 [ + + ]: 1398 : if (blockno == hscan->rs_startblock)
2242 : : {
2243 : 13 : blockno = InvalidBlockNumber;
2244 : 13 : }
2245 : : }
2246 : : }
2247 : :
2248 : 2132 : hscan->rs_cblock = blockno;
2249 : :
2250 [ + + ]: 2132 : if (!BlockNumberIsValid(blockno))
2251 : : {
2252 : 24 : hscan->rs_inited = false;
2253 : 24 : return false;
2254 : : }
2255 : :
2256 [ + - ]: 2108 : Assert(hscan->rs_cblock < hscan->rs_nblocks);
2257 : :
2258 : : /*
2259 : : * Be sure to check for interrupts at least once per page. Checks at
2260 : : * higher code levels won't be able to stop a sample scan that encounters
2261 : : * many pages' worth of consecutive dead tuples.
2262 : : */
2263 [ + - ]: 2108 : CHECK_FOR_INTERRUPTS();
2264 : :
2265 : : /* Read page using selected strategy */
2266 : 4216 : hscan->rs_cbuf = ReadBufferExtended(hscan->rs_base.rs_rd, MAIN_FORKNUM,
2267 : 2108 : blockno, RBM_NORMAL, hscan->rs_strategy);
2268 : :
2269 : : /* in pagemode, prune the page and determine visible tuple offsets */
2270 [ + + ]: 2108 : if (hscan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
2271 : 1410 : heap_prepare_pagescan(scan);
2272 : :
2273 : 2108 : hscan->rs_inited = true;
2274 : 2108 : return true;
2275 : 2132 : }
2276 : :
2277 : : static bool
2278 : 42235 : heapam_scan_sample_next_tuple(TableScanDesc scan, SampleScanState *scanstate,
2279 : : TupleTableSlot *slot)
2280 : : {
2281 : 42235 : HeapScanDesc hscan = (HeapScanDesc) scan;
2282 : 42235 : TsmRoutine *tsm = scanstate->tsmroutine;
2283 : 42235 : BlockNumber blockno = hscan->rs_cblock;
2284 : 42235 : bool pagemode = (scan->rs_flags & SO_ALLOW_PAGEMODE) != 0;
2285 : :
2286 : 42235 : Page page;
2287 : 42235 : bool all_visible;
2288 : 42235 : OffsetNumber maxoffset;
2289 : :
2290 : : /*
2291 : : * When not using pagemode, we must lock the buffer during tuple
2292 : : * visibility checks.
2293 : : */
2294 [ + + ]: 42235 : if (!pagemode)
2295 : 699 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_SHARE);
2296 : :
2297 : 42235 : page = BufferGetPage(hscan->rs_cbuf);
2298 [ + + ]: 84367 : all_visible = PageIsAllVisible(page) &&
2299 : 42132 : !scan->rs_snapshot->takenDuringRecovery;
2300 : 42235 : maxoffset = PageGetMaxOffsetNumber(page);
2301 : :
2302 : 42235 : for (;;)
2303 : : {
2304 : 42235 : OffsetNumber tupoffset;
2305 : :
2306 [ + - ]: 42235 : CHECK_FOR_INTERRUPTS();
2307 : :
2308 : : /* Ask the tablesample method which tuples to check on this page. */
2309 : 84470 : tupoffset = tsm->NextSampleTuple(scanstate,
2310 : 42235 : blockno,
2311 : 42235 : maxoffset);
2312 : :
2313 [ + + + + ]: 42235 : if (OffsetNumberIsValid(tupoffset))
2314 : : {
2315 : 40128 : ItemId itemid;
2316 : 40128 : bool visible;
2317 : 40128 : HeapTuple tuple = &(hscan->rs_ctup);
2318 : :
2319 : : /* Skip invalid tuple pointers. */
2320 : 40128 : itemid = PageGetItemId(page, tupoffset);
2321 [ + - ]: 40128 : if (!ItemIdIsNormal(itemid))
2322 : 0 : continue;
2323 : :
2324 : 40128 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, itemid);
2325 : 40128 : tuple->t_len = ItemIdGetLength(itemid);
2326 : 40128 : ItemPointerSet(&(tuple->t_self), blockno, tupoffset);
2327 : :
2328 : :
2329 [ + + ]: 40128 : if (all_visible)
2330 : 40058 : visible = true;
2331 : : else
2332 : 140 : visible = SampleHeapTupleVisible(scan, hscan->rs_cbuf,
2333 : 70 : tuple, tupoffset);
2334 : :
2335 : : /* in pagemode, heap_prepare_pagescan did this for us */
2336 [ + + ]: 40128 : if (!pagemode)
2337 : 2 : HeapCheckForSerializableConflictOut(visible, scan->rs_rd, tuple,
2338 : 1 : hscan->rs_cbuf, scan->rs_snapshot);
2339 : :
2340 : : /* Try next tuple from same page. */
2341 [ + - ]: 40128 : if (!visible)
2342 : 0 : continue;
2343 : :
2344 : : /* Found visible tuple, return it. */
2345 [ + + ]: 40128 : if (!pagemode)
2346 : 1 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
2347 : :
2348 : 40128 : ExecStoreBufferHeapTuple(tuple, slot, hscan->rs_cbuf);
2349 : :
2350 : : /* Count successfully-fetched tuples as heap fetches */
2351 [ + - + - : 40128 : pgstat_count_heap_getnext(scan->rs_rd);
# # ]
2352 : :
2353 : 40128 : return true;
2354 : 40128 : }
2355 : : else
2356 : : {
2357 : : /*
2358 : : * If we get here, it means we've exhausted the items on this page
2359 : : * and it's time to move to the next.
2360 : : */
2361 [ + + ]: 2107 : if (!pagemode)
2362 : 698 : LockBuffer(hscan->rs_cbuf, BUFFER_LOCK_UNLOCK);
2363 : :
2364 : 2107 : ExecClearTuple(slot);
2365 : 2107 : return false;
2366 : : }
2367 [ + - ]: 42235 : }
2368 : :
2369 : : Assert(0);
2370 : 42235 : }
2371 : :
2372 : :
2373 : : /* ----------------------------------------------------------------------------
2374 : : * Helper functions for the above.
2375 : : * ----------------------------------------------------------------------------
2376 : : */
2377 : :
2378 : : /*
2379 : : * Reconstruct and rewrite the given tuple
2380 : : *
2381 : : * We cannot simply copy the tuple as-is, for several reasons:
2382 : : *
2383 : : * 1. We'd like to squeeze out the values of any dropped columns, both
2384 : : * to save space and to ensure we have no corner-case failures. (It's
2385 : : * possible for example that the new table hasn't got a TOAST table
2386 : : * and so is unable to store any large values of dropped cols.)
2387 : : *
2388 : : * 2. The tuple might not even be legal for the new table; this is
2389 : : * currently only known to happen as an after-effect of ALTER TABLE
2390 : : * SET WITHOUT OIDS.
2391 : : *
2392 : : * So, we must reconstruct the tuple from component Datums.
2393 : : */
2394 : : static void
2395 : 114899 : reform_and_rewrite_tuple(HeapTuple tuple,
2396 : : Relation OldHeap, Relation NewHeap,
2397 : : Datum *values, bool *isnull, RewriteState rwstate)
2398 : : {
2399 : 114899 : TupleDesc oldTupDesc = RelationGetDescr(OldHeap);
2400 : 114899 : TupleDesc newTupDesc = RelationGetDescr(NewHeap);
2401 : 114899 : HeapTuple copiedTuple;
2402 : 114899 : int i;
2403 : :
2404 : 114899 : heap_deform_tuple(tuple, oldTupDesc, values, isnull);
2405 : :
2406 : : /* Be sure to null out any dropped columns */
2407 [ + + ]: 736462 : for (i = 0; i < newTupDesc->natts; i++)
2408 : : {
2409 [ + - ]: 621563 : if (TupleDescCompactAttr(newTupDesc, i)->attisdropped)
2410 : 0 : isnull[i] = true;
2411 : 621563 : }
2412 : :
2413 : 114899 : copiedTuple = heap_form_tuple(newTupDesc, values, isnull);
2414 : :
2415 : : /* The heap rewrite module does the rest */
2416 : 114899 : rewrite_heap_tuple(rwstate, tuple, copiedTuple);
2417 : :
2418 : 114899 : heap_freetuple(copiedTuple);
2419 : 114899 : }
2420 : :
2421 : : /*
2422 : : * Check visibility of the tuple.
2423 : : */
2424 : : static bool
2425 : 70 : SampleHeapTupleVisible(TableScanDesc scan, Buffer buffer,
2426 : : HeapTuple tuple,
2427 : : OffsetNumber tupoffset)
2428 : : {
2429 : 70 : HeapScanDesc hscan = (HeapScanDesc) scan;
2430 : :
2431 [ + + ]: 70 : if (scan->rs_flags & SO_ALLOW_PAGEMODE)
2432 : : {
2433 : 69 : uint32 start = 0,
2434 : 69 : end = hscan->rs_ntuples;
2435 : :
2436 : : /*
2437 : : * In pageatatime mode, heap_prepare_pagescan() already did visibility
2438 : : * checks, so just look at the info it left in rs_vistuples[].
2439 : : *
2440 : : * We use a binary search over the known-sorted array. Note: we could
2441 : : * save some effort if we insisted that NextSampleTuple select tuples
2442 : : * in increasing order, but it's not clear that there would be enough
2443 : : * gain to justify the restriction.
2444 : : */
2445 [ + - ]: 122 : while (start < end)
2446 : : {
2447 : 122 : uint32 mid = start + (end - start) / 2;
2448 : 122 : OffsetNumber curoffset = hscan->rs_vistuples[mid];
2449 : :
2450 [ + + ]: 122 : if (tupoffset == curoffset)
2451 : 69 : return true;
2452 [ + + ]: 53 : else if (tupoffset < curoffset)
2453 : 26 : end = mid;
2454 : : else
2455 : 27 : start = mid + 1;
2456 [ + + ]: 122 : }
2457 : :
2458 : 0 : return false;
2459 : 69 : }
2460 : : else
2461 : : {
2462 : : /* Otherwise, we have to check the tuple individually. */
2463 : 2 : return HeapTupleSatisfiesVisibility(tuple, scan->rs_snapshot,
2464 : 1 : buffer);
2465 : : }
2466 : 70 : }
2467 : :
2468 : : /*
2469 : : * Helper function get the next block of a bitmap heap scan. Returns true when
2470 : : * it got the next block and saved it in the scan descriptor and false when
2471 : : * the bitmap and or relation are exhausted.
2472 : : */
2473 : : static bool
2474 : 44425 : BitmapHeapScanNextBlock(TableScanDesc scan,
2475 : : bool *recheck,
2476 : : uint64 *lossy_pages, uint64 *exact_pages)
2477 : : {
2478 : 44425 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) scan;
2479 : 44425 : HeapScanDesc hscan = (HeapScanDesc) bscan;
2480 : 44425 : BlockNumber block;
2481 : 44425 : void *per_buffer_data;
2482 : 44425 : Buffer buffer;
2483 : 44425 : Snapshot snapshot;
2484 : 44425 : int ntup;
2485 : 44425 : TBMIterateResult *tbmres;
2486 : 44425 : OffsetNumber offsets[TBM_MAX_TUPLES_PER_PAGE];
2487 : 44425 : int noffsets = -1;
2488 : :
2489 [ + - ]: 44425 : Assert(scan->rs_flags & SO_TYPE_BITMAPSCAN);
2490 [ + - ]: 44425 : Assert(hscan->rs_read_stream);
2491 : :
2492 : 44425 : hscan->rs_cindex = 0;
2493 : 44425 : hscan->rs_ntuples = 0;
2494 : :
2495 : : /* Release buffer containing previous block. */
2496 [ + + ]: 44425 : if (BufferIsValid(hscan->rs_cbuf))
2497 : : {
2498 : 42414 : ReleaseBuffer(hscan->rs_cbuf);
2499 : 42414 : hscan->rs_cbuf = InvalidBuffer;
2500 : 42414 : }
2501 : :
2502 : 44425 : hscan->rs_cbuf = read_stream_next_buffer(hscan->rs_read_stream,
2503 : : &per_buffer_data);
2504 : :
2505 [ + + ]: 44425 : if (BufferIsInvalid(hscan->rs_cbuf))
2506 : : {
2507 : : /* the bitmap is exhausted */
2508 : 1952 : return false;
2509 : : }
2510 : :
2511 [ + - ]: 42473 : Assert(per_buffer_data);
2512 : :
2513 : 42473 : tbmres = per_buffer_data;
2514 : :
2515 [ + - ]: 42473 : Assert(BlockNumberIsValid(tbmres->blockno));
2516 [ + - ]: 42473 : Assert(BufferGetBlockNumber(hscan->rs_cbuf) == tbmres->blockno);
2517 : :
2518 : : /* Exact pages need their tuple offsets extracted. */
2519 [ + + ]: 42473 : if (!tbmres->lossy)
2520 : 15743 : noffsets = tbm_extract_page_tuple(tbmres, offsets,
2521 : : TBM_MAX_TUPLES_PER_PAGE);
2522 : :
2523 : 42473 : *recheck = tbmres->recheck;
2524 : :
2525 : 42473 : block = hscan->rs_cblock = tbmres->blockno;
2526 : 42473 : buffer = hscan->rs_cbuf;
2527 : 42473 : snapshot = scan->rs_snapshot;
2528 : :
2529 : 42473 : ntup = 0;
2530 : :
2531 : : /*
2532 : : * Prune and repair fragmentation for the whole page, if possible.
2533 : : */
2534 : 42473 : heap_page_prune_opt(scan->rs_rd, buffer);
2535 : :
2536 : : /*
2537 : : * We must hold share lock on the buffer content while examining tuple
2538 : : * visibility. Afterwards, however, the tuples we have found to be
2539 : : * visible are guaranteed good as long as we hold the buffer pin.
2540 : : */
2541 : 42473 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
2542 : :
2543 : : /*
2544 : : * We need two separate strategies for lossy and non-lossy cases.
2545 : : */
2546 [ + + ]: 42473 : if (!tbmres->lossy)
2547 : : {
2548 : : /*
2549 : : * Bitmap is non-lossy, so we just look through the offsets listed in
2550 : : * tbmres; but we have to follow any HOT chain starting at each such
2551 : : * offset.
2552 : : */
2553 : 15743 : int curslot;
2554 : :
2555 : : /* We must have extracted the tuple offsets by now */
2556 [ + - ]: 15743 : Assert(noffsets > -1);
2557 : :
2558 [ + + ]: 511696 : for (curslot = 0; curslot < noffsets; curslot++)
2559 : : {
2560 : 495953 : OffsetNumber offnum = offsets[curslot];
2561 : 495953 : ItemPointerData tid;
2562 : 495953 : HeapTupleData heapTuple;
2563 : :
2564 : 495953 : ItemPointerSet(&tid, block, offnum);
2565 [ + + ]: 495953 : if (heap_hot_search_buffer(&tid, scan->rs_rd, buffer, snapshot,
2566 : : &heapTuple, NULL, true))
2567 : 472528 : hscan->rs_vistuples[ntup++] = ItemPointerGetOffsetNumber(&tid);
2568 : 495953 : }
2569 : 15743 : }
2570 : : else
2571 : : {
2572 : : /*
2573 : : * Bitmap is lossy, so we must examine each line pointer on the page.
2574 : : * But we can ignore HOT chains, since we'll check each tuple anyway.
2575 : : */
2576 : 26730 : Page page = BufferGetPage(buffer);
2577 : 26730 : OffsetNumber maxoff = PageGetMaxOffsetNumber(page);
2578 : 26730 : OffsetNumber offnum;
2579 : :
2580 [ + + ]: 230745 : for (offnum = FirstOffsetNumber; offnum <= maxoff; offnum = OffsetNumberNext(offnum))
2581 : : {
2582 : 204015 : ItemId lp;
2583 : 204015 : HeapTupleData loctup;
2584 : 204015 : bool valid;
2585 : :
2586 : 204015 : lp = PageGetItemId(page, offnum);
2587 [ + - ]: 204015 : if (!ItemIdIsNormal(lp))
2588 : 0 : continue;
2589 : 204015 : loctup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2590 : 204015 : loctup.t_len = ItemIdGetLength(lp);
2591 : 204015 : loctup.t_tableOid = scan->rs_rd->rd_id;
2592 : 204015 : ItemPointerSet(&loctup.t_self, block, offnum);
2593 : 204015 : valid = HeapTupleSatisfiesVisibility(&loctup, snapshot, buffer);
2594 [ + + ]: 204015 : if (valid)
2595 : : {
2596 : 203994 : hscan->rs_vistuples[ntup++] = offnum;
2597 : 407988 : PredicateLockTID(scan->rs_rd, &loctup.t_self, snapshot,
2598 : 203994 : HeapTupleHeaderGetXmin(loctup.t_data));
2599 : 203994 : }
2600 : 408030 : HeapCheckForSerializableConflictOut(valid, scan->rs_rd, &loctup,
2601 : 204015 : buffer, snapshot);
2602 [ - - + ]: 204015 : }
2603 : 26730 : }
2604 : :
2605 : 42473 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2606 : :
2607 [ + - ]: 42473 : Assert(ntup <= MaxHeapTuplesPerPage);
2608 : 42473 : hscan->rs_ntuples = ntup;
2609 : :
2610 [ + + ]: 42473 : if (tbmres->lossy)
2611 : 26730 : (*lossy_pages)++;
2612 : : else
2613 : 15743 : (*exact_pages)++;
2614 : :
2615 : : /*
2616 : : * Return true to indicate that a valid block was found and the bitmap is
2617 : : * not exhausted. If there are no visible tuples on this page,
2618 : : * hscan->rs_ntuples will be 0 and heapam_scan_bitmap_next_tuple() will
2619 : : * return false returning control to this function to advance to the next
2620 : : * block in the bitmap.
2621 : : */
2622 : 42473 : return true;
2623 : 44425 : }
2624 : :
2625 : : /* ------------------------------------------------------------------------
2626 : : * Definition of the heap table access method.
2627 : : * ------------------------------------------------------------------------
2628 : : */
2629 : :
2630 : : static const TableAmRoutine heapam_methods = {
2631 : : .type = T_TableAmRoutine,
2632 : :
2633 : : .slot_callbacks = heapam_slot_callbacks,
2634 : :
2635 : : .scan_begin = heap_beginscan,
2636 : : .scan_end = heap_endscan,
2637 : : .scan_rescan = heap_rescan,
2638 : : .scan_getnextslot = heap_getnextslot,
2639 : :
2640 : : .scan_set_tidrange = heap_set_tidrange,
2641 : : .scan_getnextslot_tidrange = heap_getnextslot_tidrange,
2642 : :
2643 : : .parallelscan_estimate = table_block_parallelscan_estimate,
2644 : : .parallelscan_initialize = table_block_parallelscan_initialize,
2645 : : .parallelscan_reinitialize = table_block_parallelscan_reinitialize,
2646 : :
2647 : : .index_fetch_begin = heapam_index_fetch_begin,
2648 : : .index_fetch_reset = heapam_index_fetch_reset,
2649 : : .index_fetch_end = heapam_index_fetch_end,
2650 : : .index_fetch_tuple = heapam_index_fetch_tuple,
2651 : :
2652 : : .tuple_insert = heapam_tuple_insert,
2653 : : .tuple_insert_speculative = heapam_tuple_insert_speculative,
2654 : : .tuple_complete_speculative = heapam_tuple_complete_speculative,
2655 : : .multi_insert = heap_multi_insert,
2656 : : .tuple_delete = heapam_tuple_delete,
2657 : : .tuple_update = heapam_tuple_update,
2658 : : .tuple_lock = heapam_tuple_lock,
2659 : :
2660 : : .tuple_fetch_row_version = heapam_fetch_row_version,
2661 : : .tuple_get_latest_tid = heap_get_latest_tid,
2662 : : .tuple_tid_valid = heapam_tuple_tid_valid,
2663 : : .tuple_satisfies_snapshot = heapam_tuple_satisfies_snapshot,
2664 : : .index_delete_tuples = heap_index_delete_tuples,
2665 : :
2666 : : .relation_set_new_filelocator = heapam_relation_set_new_filelocator,
2667 : : .relation_nontransactional_truncate = heapam_relation_nontransactional_truncate,
2668 : : .relation_copy_data = heapam_relation_copy_data,
2669 : : .relation_copy_for_cluster = heapam_relation_copy_for_cluster,
2670 : : .relation_vacuum = heap_vacuum_rel,
2671 : : .scan_analyze_next_block = heapam_scan_analyze_next_block,
2672 : : .scan_analyze_next_tuple = heapam_scan_analyze_next_tuple,
2673 : : .index_build_range_scan = heapam_index_build_range_scan,
2674 : : .index_validate_scan = heapam_index_validate_scan,
2675 : :
2676 : : .relation_size = table_block_relation_size,
2677 : : .relation_needs_toast_table = heapam_relation_needs_toast_table,
2678 : : .relation_toast_am = heapam_relation_toast_am,
2679 : : .relation_fetch_toast_slice = heap_fetch_toast_slice,
2680 : :
2681 : : .relation_estimate_size = heapam_estimate_rel_size,
2682 : :
2683 : : .scan_bitmap_next_tuple = heapam_scan_bitmap_next_tuple,
2684 : : .scan_sample_next_block = heapam_scan_sample_next_block,
2685 : : .scan_sample_next_tuple = heapam_scan_sample_next_tuple
2686 : : };
2687 : :
2688 : :
2689 : : const TableAmRoutine *
2690 : 1670229 : GetHeapamTableAmRoutine(void)
2691 : : {
2692 : 1670229 : return &heapam_methods;
2693 : : }
2694 : :
2695 : : Datum
2696 : 99347 : heap_tableam_handler(PG_FUNCTION_ARGS)
2697 : : {
2698 : 99347 : PG_RETURN_POINTER(&heapam_methods);
2699 : : }
|
